WO1998047105A1 - Appareil et procede pour la visualisation de volume en temps reel, en parallele et en perspective - Google Patents

Appareil et procede pour la visualisation de volume en temps reel, en parallele et en perspective Download PDF

Info

Publication number
WO1998047105A1
WO1998047105A1 PCT/US1998/007405 US9807405W WO9847105A1 WO 1998047105 A1 WO1998047105 A1 WO 1998047105A1 US 9807405 W US9807405 W US 9807405W WO 9847105 A1 WO9847105 A1 WO 9847105A1
Authority
WO
WIPO (PCT)
Prior art keywords
slice
buffer
ray
compositing
inte
Prior art date
Application number
PCT/US1998/007405
Other languages
English (en)
Inventor
Arie E. Kaufman
Ingmar Bitter
Original Assignee
The Research Foundation Of State University Of New York
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from US08/910,575 external-priority patent/US5847711A/en
Application filed by The Research Foundation Of State University Of New York filed Critical The Research Foundation Of State University Of New York
Priority to JP10525927A priority Critical patent/JPH11508386A/ja
Priority to EP98918164A priority patent/EP0992023A4/fr
Priority to AU71139/98A priority patent/AU732652B2/en
Priority to CA002285966A priority patent/CA2285966A1/fr
Publication of WO1998047105A1 publication Critical patent/WO1998047105A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T15/00—3D [Three Dimensional] image rendering
    • G06T15/10—Geometric effects

Definitions

  • the present invention relates to three dimensional (3-D) graphics and volumetric imaging, and more particularly to an apparatus and method for real-time parallel and perspective projection of high resolution volumetric images.
  • Image rendering is the process of converting complex information to a format which is amendable to human understanding while maintaining the integrity and accuracy of the information.
  • Volumetric data which consists of information relating to three-dimensional phenomena, is one species of complex information that can benefit from improved image rendering techniques.
  • the process of analyzing volumetric data to determine, from a given viewpoint, which portions of a volume are to be presented is commonly referred to as volume visualization.
  • Traditional methods of volume visualization operate by scanning through data points in a sequential manner in order to provide an accurate representation of an object. The need to model objects in real-time and the advantage of doing so using computer graphic systems is clear.
  • the volume visualization system 1 includes a cubic frame buffer 2 having a skewed memory organization which enables conflict free access of a beam of voxels in any orthographic direction, a two-dimensional (2-D) skewed buffer 4, a ray projection tree 6, and two conveyers 8, 10.
  • the conveyors are commonly referred to as barrel shifters.
  • a first conveyor 8 is coupled between the cubic frame buffer and the two dimensional skewed buffer, while a second conveyor 10 is coupled between the two-dimensional skewed buffer and the ray projection tree.
  • This volume visualization system is capable of providing parallel projection in O(n 2 logn) time, where n is the measurement of one axis of the cubic frame buffer.
  • the traditional volume visualization system 1 operates by casting viewing rays 12, originating at a pixel in a projection plane (not shown), through the cubic frame buffer 2 along a selected viewing direction.
  • the viewing rays access a plurality of voxels 14 (defining a projection ray plane (PRP)16) stored in the cubic frame buffer.
  • the voxels defining the PRP are simultaneously retrieved by orthogonal beams 18 and provided to the conveyor 8.
  • the conveyor 8 provides a 2-D shearing of the voxels of the orthogonal beams which define the PRP.
  • This 2-D shearing serves to align all of the voxels of each discrete viewing ray along a direction parallel to a 2-D axis of the 2-D skewed buffer to provide skewed viewing rays. Once the viewing rays are aligned in the 2-D skewed buffer, the skewed viewing rays can be retrieved and processed by the ray projection tree 6.
  • the accessed skewed viewing rays are provided to conveyor 10.
  • the conveyor 10 preforms a deskewing operation in order to match the physical sequential order of the input modules of the ray projection tree 6 to the sequential order of the voxels of each viewing ray. Specifically, each viewing ray is shifted such that the first voxel in each projection ray appears at the corresponding first input position of the ray projection tree.
  • the voxels of each ray are then processed by the ray projection tree in parallel so as to generate a pixel value associated with that projection ray.
  • the speed at which the system operates is limited by the system architecture which provides arbitrary parallel and orthographic projections in O(n 2 log n) time.
  • the ray projection tree requires that each projection ray be provided thereto in a specific orientation. This requires a conveyor between the two-dimensional skewed buffer and the ray projection tree which adds to the overall hardware required by the system and the time needed for volume visualization.
  • the traditional system provides only surface approximations of discrete projection rays by utilizing the closest non-transparent discrete voxel to points along the discrete projection rays instead of actual values along the continuous projection rays. This provides a somewhat inaccurate representation of the object.
  • the conveyors are not readily capable of shifting data in a manner required for perspective projection (fanning and defanning of data) and real-time visualization of four-dimensional (4-D) data.
  • a first step consists of selecting viewing and processing parameters which define a view point from which the image is generated and at least one base plane of a three dimensional buffer in which discrete voxels having a location and at least one voxel value are stored.
  • the base plane is employed for projection purposes.
  • the viewing and processing parameters also define first and last processing slices of the three dimensional buffer.
  • the method includes the additional steps of initializing a compositing buffer having a plurality of pixels, each of which has at least color, transparency and position associated with it; sampling voxel values from the three dimensional buffer onto a current slice of sample points which are parallel to the first processing slice, in order to provide sample point values; and combining the sample point values with the pixels of the compositing buffer.
  • the current slice is the first processing slice during the first execution of the sampling step.
  • the combination occurs along a plurality of interslice ray segments which extend only between the current slice and an adjacent slice associated with the compositing buffer.
  • the sampling and combining steps are repeated by sequentially sweeping through subsequent slices of sample points which are parallel to the first processing slice, until the last processing slice is reached. Each of the subsequent slices in turn becomes the current slice.
  • the sampling step includes sampling voxel values from the three dimensional buffer directly onto grid points of the current slice of sample points, with the sample points coinciding with the grid points.
  • the combining step includes the substep of casting a plurality of sight rays from the view point through the grid points of the current slice of sample points.
  • the interslice ray segments extend along the sight rays from the sample points of the current slice to the adjacent slice associated with the compositing buffer, which they intersect at a plurality of intersection points. Further, when the sampling and combining steps are repeated, the sight rays are recast as each of the subsequent slices becomes the current slice.
  • an additional step includes performing one of zero order, first order, second order, third order, and higher order interpolation, or an adaptive combination thereof, among the pixels of the compositing buffer adjacent the intersection points in order to provide interpolated pixel values for combination with the sample points of the current slice.
  • the combining step includes the sub-step of casting a plurality of sight rays from the view point through locations in the adjacent slice associated with the compositing buffer which correspond to grid point locations of the pixels within the compositing buffer.
  • the sight rays intersect the current slice at a plurality of intersection points.
  • the sampling step includes sampling the voxel values from the three dimensional buffer onto the plurality of intersection points by performing zero, first, second, third or higher order interpolation, or an adaptive combination thereof, among voxel values associated with the voxel locations adjacent the plurality of intersection points.
  • the sampling step includes sampling the voxel values from the three dimensional buffer directly onto grid points of the current slice of sample points, with the sample points coinciding with the grid points.
  • a first execution of the combining step includes the sub-step of casting a plurality of sight rays from the view point through the grid points of the current slice of sample points, with the interslice ray segments extending from the sample points of the current slice to intersections of the sight rays with the compositing buffer.
  • the method further includes performing one of zero, first, second, third or higher order interpolation, or an adaptive combination thereof, among pixels of the compositing buffer adjacent to the intersection points in order to provide interpolated pixel values for combination with the sample points of the current slice.
  • the results of the interpolation and the combination step are stored, with an accumulated offset, in grid point locations of those pixels in the compositing buffer which are closest to the intersections of the sight rays with the compositing buffer.
  • merging and splitting of the interslice ray segments is performed. Merging is performed when an attempt is made to store two or more of the results of the interpolation and combination at an identical grid point location; whereas splitting is performed when interslice ray segments diverge beyond a divergence threshold.
  • Additional steps in this preferred method include maintaining the sight rays formed in the first execution of the combining step throughout the sequential sweeping except when the merging or splitting is performed; generating a new sight ray from the view point to a preselected location in the neighborhood of the identical grid point location when the merging is performed; generating two new sight rays from the view point to corresponding preselected locations in the current slice when splitting is performed; and storing the results of the interpolation and the combination in an adjacent grid point location when the accumulated offset exceeds a predetermined value. In the latter case, the storing is done together with a new value for the accumulated offset.
  • an additional step includes dividing the three dimensional buffer into a plurality of zones, separated by zone boundaries, with the merging and splitting being confined to the zones and not occurring across the zone boundaries.
  • the zones and zone boundaries are selected to enhance the image sharpness and accuracy; most preferably by forming generalized three dimensional pyramids extending from the view point through boundaries of pixels associated with the base plane into the three dimensional buffer.
  • a first execution of the combining step includes the substep of casting a plurality of sight rays from the view point through locations in the adjacent slice associated with the compositing buffer.
  • the locations correspond to grid point locations of the pixels within the compositing buffer and the sight rays intersect the current slice at a plurality of intersection points.
  • the interslice ray segments extend from the grid point locations of the pixels within the compositing buffer to the intersection points.
  • the sampling step includes sampling the voxel values from the 3-D buffer onto the plurality of intersection points by one of zero order, first order, second order, third order and higher order interpolation, and adaptive combinations thereof, performed among voxel values associated with voxel locations adjacent the plurality of intersection points. Additional steps include storing the results of the interpolation and the combination, with an accumulated offset, in the grid point locations of the pixels within the compositing buffer; performing one of merging and splitting of the interslice ray segments; and maintaining the sight rays formed in the first execution of the combining step, except when one of the merging and splitting is performed.
  • the merging is performed when an attempt is made to store two or more of the results of the interpolation and the combination in an identical grid point location. Two or more of the interslice slice ray segments are merged in this case.
  • the splitting is performed for at least one interslice ray segment when adjacent interslice ray segments diverge beyond a divergence threshold.
  • the fourth preferred method includes the additional steps of generating a new sight ray from the view point to a preselected location in the neighborhood of the identical grid point location when the merging is performed; generating two new sight rays from the view point to the corresponding preselected locations in the compositing buffer when the splitting is performed; and storing the results of the interpolation and the combination in an adjacent one of the grid point locations when the accumulated offset exceeds a predetermined value. Storing is done together with a new value of the accumulated offset.
  • the method further includes the additional step of dividing the three dimensional buffer into a plurality of zones which are separated by zone boundaries, with the merging and splitting being confined to individual zones and adjacent zones on either side of that zone and not occurring, for a given zone, across more than one of the zone boundaries on each side of the given zone.
  • the zones and zone boundaries are selected for enhancing image sharpness and accuracy, and for reduction of aliasing. They are preferably determined by forming generalized three dimensional pyramids extending from the view point through boundaries of the pixels associated with the base plane into the three dimensional buffer.
  • An apparatus, in accordance with the present invention, for parallel and perspective real-time volume visualization via ray-slice-sweeping includes a three dimensional buffer; a pixel bus; a plurality of rendering pipelines; and a control unit.
  • the apparatus is responsive to viewing and processing parameters which define the view point, and the apparatus generates a three dimensional volume projection image from the view point.
  • the image has a plurality of pixels.
  • the three dimensional buffer stores a plurality of discrete voxels, each of which has a location and at least one voxel value associated therewith.
  • the three dimensional buffer includes a plurality of memory units and the viewing and processing parameters define at least one base plane of the three dimensional buffer as well as first and last processing slices of the three dimensional buffer.
  • the pixel bus provides global horizontal communication.
  • the control unit initially designates the first processing slice as a current slice and controls sweeping through subsequent slices of the three dimensional buffer as current slices, until the last processing slice is reached.
  • Each of the plurality of rendering pipelines is vertically coupled to a corresponding one of the plurality of memory units and to the pixel bus, and each of the rendering pipelines has horizontal communication with at most its two nearest neighbors.
  • Each of the rendering pipelines in turn comprises at least a first slice unit; a compositing unit; a two-dimensional slice compositing buffer; and a first bilinear interpolation unit.
  • the slice unit has an input coupled to the corresponding one of the plurality of memory units and an output, and it includes a current slice of sample points which are parallel to the first processing slice.
  • the slice unit receives voxel values from the three dimensional buffer onto sample points in order to provide sample point values.
  • the compositing unit has an input coupled to the output of the slice unit and has an output coupled to the pixel bus.
  • the two dimensional slice compositing buffer has a plurality of pixels, each of which has at least color, transparency and position associated with it.
  • the compositing buffer has an input coupled to the output of the compositing unit and has an output coupled to the input of the compositing unit.
  • the first bilinear interpolation unit has an input which is coupled to either the output of the compositing buffer or the corresponding one of the plurality of memory units.
  • the bilinear inte ⁇ olation unit also has an output.
  • the output of the bilinear interpolation unit is coupled to the input of the compositing unit when the input of the bilinear interpolation unit is coupled to the output of the compositing buffer.
  • the output of the bilinear interpolation unit is coupled to the input of the slice unit when the input of the bilinear interpolation unit is coupled to the corresponding one of the plurality of memory units.
  • the sample point values are combined with the pixels of the compositing buffer in the compositing unit.
  • the combination occurs along a plurality of interslice race segments which extend only between the current slice in the slice unit and a slice contained in the two-dimensional slice compositing buffer.
  • the slice unit receives the voxel values from the three-dimensional buffer directly onto grid points of the current slice of sample points, and the sample points coincide with the grid points.
  • the interslice ray segments extend along the plurality of sight rays cast from the view point through the grid points of the current slice of sample points.
  • the interslice ray segments extend from the sample points of the current slice in the slice unit to the slice contained in the two-dimensional slice compositing buffer, and they intersect the slice contained in the two-dimensional slice compositing buffer at a plurality of intersection points.
  • the bilinear interpolation unit has its input coupled to the output of the compositing buffer and its output coupled to the input of the compositing unit. It receives signals associated with pixels of the two-dimensional slice compositing buffer which are adjacent to the intersection points and provides interpolated pixel values for combination with the sample points of the current slice in the compositing unit.
  • the control unit recasts the sight rays as each of the subsequent slices becomes the current slice.
  • the first bilinear inte ⁇ olation unit has its input coupled to the three dimensional buffer and its output coupled to the input of the slice unit.
  • the interslice ray segments extend along a plurality of sight rays cast from the view point through locations in the slice contained in the two dimensional slice compositing buffer. The locations correspond to grid point locations of the pfxels within the compositing buffer, and the sight rays intersect the current slice at a plurality of intersection points.
  • the first bilinear inte ⁇ olation unit receives voxel values associated with voxel locations adjacent the plurality of intersection points and inte ⁇ olates among the voxel values to provide inte ⁇ olated voxel values associated with the plurality of intersection points.
  • the control unit recasts the sight rays as each of the subsequent slices becomes the current slice.
  • the slice unit receives the voxel values from the three dimensional buffer directly onto grid points of the current slice of sample points. The sample points coincide with the grid points.
  • the interslice ray segments extend along a plurality of sight rays cast from the view point through the grid points of the current slice of sample points.
  • the interslice ray segments extend from the sample points of the current slice in the slice unit to the slice contained in the two dimensional slice compositing buffer.
  • the interslice ray segments intersect the slice contained in the two dimensional slice compositing buffer at a plurality of intersection points.
  • the first bilinear inte ⁇ olation unit has its input coupled to the output of the compositing buffer and its output coupled to the input of the compositing unit.
  • the bilinear inte ⁇ olation unit receives signals associated with the pixels of the two dimensional slice compositing buffer which are adjacent to the intersection points and provides inte ⁇ olated pixel values for combination with the sample points of the current slice in the compositing unit.
  • results of the combination of the inte ⁇ olated pixel values and the sample points of the current slice are stored, with an accumulated offset, in grid point locations of those of the pixels in the compositing buffer which are closest to the intersection of the sight rays with the compositing buffer.
  • the control unit enables one of merging and splitting of the interslice ray segments. Merging of two or more interslice ray segments is performed when an attempt is made to store two or more of the results of the inte ⁇ olation and the combination at an identical grid point location. Splitting of at least one of the interslice ray segments is performed when adjacent interslice ray segments diverge beyond a divergence threshold.
  • the control unit maintains the sight rays throughout the sequential sweeping through the subsequent slices, except when either merging or splitting is performed. When the merging is performed, the control unit generates a new sight ray from the view point to a preselected location in the neighborhood of the identical grid point location.
  • the control unit When the splitting is performed, the control unit generates two new sight rays from the view point to corresponding preselected locations in the current slice.
  • the control unit stores the results of the inte ⁇ olation and the combination in an adjacent one of the grid point locations when the accumulated offset exceeds a predetermined value. In this case, the storing is done together with a new accumulated offset value.
  • the control unit divides the three dimensional buffer into a plurality of zones separated by zone boundaries, with merging and splitting confined to the zones and not occurring across the zone boundaries. The zones and zone boundaries are selected for enhancing the image sha ⁇ ness and accuracy.
  • the first bilinear inte ⁇ olation unit has its input coupled to the three dimensional buffer and its output coupled to the input of the slice unit.
  • the interslice ray segments extend along a plurality of sight rays cast from the view point through locations in the slice contained in the two dimensional slice compositing buffer. The locations correspond to grid point locations of the pixels within the compositing buffer. The sight rays intersect the current slice at a plurality of intersection points.
  • the first bilinear inte ⁇ olation unit receives voxel values associated with voxel locations adjacent the plurality of intersection points and inte ⁇ olates among the voxel values to provide inte ⁇ olated voxel values associated with the plurality of intersection points for combination with the pixels of the compositing buffer. Results of the combination are stored, with an accumulated offset, in the grid point locations of the pixels within the compositing buffer.
  • the control units enables one of merging and splitting of the interslice ray segments. Merging of two or more interslice ray segments is performed when an attempt is made to store two or more of the results of the inte ⁇ olation and the combination at an identical grid point location. Splitting is performed for at least one of the interslice ray segments when adjacent interslice ray segments diverge beyond a divergence threshold.
  • the control unit maintains the sight rays throughout the sequential sweeping through the subsequent slices, except when merging or splitting is performed.
  • control unit stores the results of the inte ⁇ olation and the combination in an adjacent one of the grid point locations when the accumulated offset exceeds a predetermined value. The storing is done together with a new value for the accumulated offset.
  • control unit divides the three dimensional buffer into a plurality of zones separated by zone boundaries, with the merging and splitting confined to individual zones and adjacent zones only and not occurring, for any given zone, across more than one of the zone boundaries on each side of the given zone.
  • the zones and zone boundaries are selected for enhancing image sha ⁇ ness and accuracy, and for reduction of aliasing.
  • the apparatus of the present invention can also employ a global feedback connection between the pixel bus and the three dimensional buffer.
  • the global feedback connection forms a global loop which is configured for feedback from the pixel bus to the 3-D buffer, and subsequently to any intermediate stage in the plurality of rendering pipelines.
  • the global feedback loop can be used with any of the foregoing embodiments of the apparatus of the present invention.
  • the apparatus and method of the present invention su ⁇ asses existing 3-D voxel based graphics methods and architectures in terms of performance, simplicity, image quality, expandability and ease of hardware realization, as well as low overhead, so as to provide real-time high resolution parallel and perspective volume viewing from any arbitrary direction.
  • a preferred form of the apparatus and method for real-time volume visualization, as well as other embodiments, objects, features and advantages of this invention, will be apparent from the following detailed description of illustrative embodiments thereof, which is to be read in connection with the accompanying drawings.
  • Fig. 1 is a block diagram of a traditional volume visualization system
  • Fig. 2 is a representation showing the inter-relationship of a cubic frame buffer, viewing rays, orthogonal beams, projection ray planes and the two-dimensional skewed buffers of the traditional volume visualization system
  • Fig. 3 is a functional block diagram of the apparatus for providing a 3-D volume projection of an object from a desired viewing direction constructed in a accordance with the present invention
  • Fig. 4 is a functional block diagram of an interconnection mechanism which couples a cubic frame buffer and a two dimensional buffer
  • Fig. 5A is a graph of a 10-neighborhood gradient estimation method in accordance with the present invention
  • Fig. 5B is a graph of a 26-neighborhood gradient estimation method is accordance with the present invention.
  • Fig. 5C is a graph of a 12-neighborhood gradient estimation method in accordance with the present invention.
  • Fig. 5D is a graph of an 8 -neighborhood gradient estimation method in accordance with the present invention
  • Fig 6. is a functional block diagram of a projection mechanism constructed in accordance with the present invention
  • Fig. 7A is a graph of a first method of inte ⁇ olation utilized in accordance with the present invention
  • Fig. 7B is a graph of a second method of inte ⁇ olation utilized in accordance with the present invention
  • Fig. 8A is a diagram of the method of inte ⁇ olation for a parallel projection
  • Fig. 8B is a diagram of the method of inte ⁇ olation for perspective projection
  • Fig. 8C is a diagram of the method of inte ⁇ olation for a parallel projection
  • Fig. 8D is a diagram of the method of inte ⁇ olation for a perspective projection
  • Fig. 9 is a diagram of a modified method for inte ⁇ olation in accordance with the present invention.
  • Fig. 10 is a diagram of the maximum offset estimation for use in a method of inte ⁇ olation in accordance with the present invention
  • Fig. 11 is a diagram showing the sampling rates of the cubic frame buffer based upon differing fields of view;
  • Fig. 12 is a functional block diagram of an alternative form of the apparatus for providing a 3-D volume projection of an object from a desired viewing direction constructed in accordance with the present invention
  • Fig. 13 is a functional diagram showing a voxel beam and a slice of voxel data as part of a cubic frame buffer;
  • Fig. 14 is a functional diagram of the method of the present invention.
  • Fig. 15 is a functional diagram showing the skewing scheme of voxel data for adjacent voxel beams and projection ray planes;
  • Fig. 16 is a functional block diagram of the apparatus for providing a 3-D volume projection of an object in accordance with the present invention;
  • Fig. 17 shows a ray-slice-sweeping method of the present invention with front- to-back compositing
  • Fig. 18 shows a ray-slice-sweeping method of the present invention with back- to-front compositing
  • Fig. 19 shows compositing techniques of the present invention for front-to-back and back-to-front cases
  • Fig. 20 shows an overview of the architecture of a preferred form of the present invention
  • Fig. 21 depicts partial beam processing order as employed in the present invention
  • Fig. 22 depicts an individual rendering pipeline of a preferred form of apparatus in accordance with the present invention
  • Fig. 23 defines a sight ray and different coordinate systems employed in the present invention
  • Fig. 24 illustrates gradient components and the unskewed spacial relationship of six computation parameters as employed in gradient estimation with the present invention
  • Fig. 25 shows the skewed position of the gradient computation parameters depicted in Fig. 24;
  • Fig. 26 shows a signal flow graph for the gradient y- component of the present invention, including de-skewing and proper time lineup;
  • Fig. 27 shows a signal flow graph for the gradient y- component in the present invention, with only nearest neighbor connections
  • Fig. 28 shows the a signal flow graph of the gradient y- component in a form of the present invention wherein the result is computed in the pipeline of the corresponding center voxel;
  • Fig. 29 shows two consecutive full beams of the present invention in skewed space
  • Fig. 30 shows two consecutive full beams of the present invention in time
  • Fig. 31 shows, in signal flow graph form, another form of the gradient y- component of the present invention.
  • Fig. 32 shows a signal flow graph for all gradient components in a configuration which minimizes pin count
  • Fig. 33 shows a signal flow graph for all gradient components in a configuration which minimizes internal buffer size
  • Fig. 34 shows a compositing method in accordance with the present invention employing a compositing buffer and bilinear inte ⁇ olation
  • Fig. 35 depicts the skewed relative positions of a current voxel and voxels in a previous slice
  • Fig. 36 depicts the functionality of the pipeline stages in bilinear inte ⁇ olation for back-to-front compositing
  • Fig. 37 depicts a signal flow graph for bilinear inte ⁇ olation in accordance with the present invention
  • Fig. 38 compares the voxel energy distribution in the ray-slice-sweeping method of the present invention to that in traditional ray casting techniques;
  • Fig. 39 depicts slice-order processing in the present invention
  • Fig. 40 depicts a new type of ray-casting in accordance with the present invention, known as slice-order ray-casting;
  • Fig. 41 depicts a two dimensional example of perspective ray-merging in accordance with the present invention.
  • Fig. 42 depicts the unit distance energy distribution in a slice-order energy- splatting method in accordance with the present invention
  • Fig. 43 depicts the ray-slice-sweeping method of the present invention applied to a three dimensional volume of voxel elements
  • Fig. 44 depicts a form of the present invention wherein inte ⁇ olation is performed on the current slice
  • Fig. 45 depicts a form of the present invention wherein the voxel volume is divided into a plurality of zones
  • Fig. 46 provides details of a preferred form of apparatus in accordance with the present invention.
  • Fig. 47 depicts a method according to the present invention employing ray splitting with inte ⁇ olation in the compositing buffer.
  • Fig. 48 depicts a method according to the present invention employing ray splitting with inte ⁇ olation on the current slice.
  • the method and apparatus of the present invention are capable of manipulating data and supporting real-time visualization of high resolution voxel-based data sets.
  • the method and apparatus are designed for use as a voxel-based system as described in the issued patents and pending applications of Arie Kaufman, a named inventor of this application, including "Method Of Converting Continuous Three- Dimensional Geometrical Representations Into Discrete Three- Dimensional Voxel-Based Representations Within A Three-Dimensional Voxel-Based System", which issued an August 6, 1991 , as U.S. Patent No.
  • Patent 5,101,475 Metal And Apparatus For Real-Time Volume Rendering From An Arbitrary Viewing Direction
  • U.S. Serial No. 08/097,637 Metal And Apparatus For Generating Realistic Images Using A Discrete Representation
  • U.S. Serial No. 07/855,223 the entire disclosure of each of these references is inco ⁇ orated herein by reference.
  • the apparatus of the present invention 20 preferably includes six basic components. These include a cubic frame buffer 22 having a plurality of memory storage units capable of storing voxels therein, three two-dimensional (2-D) buffers 24 and an interconnection mechanism 26 coupling the cubic frame buffer to each of the 2-D buffers.
  • the cubic frame buffer is a three-dimensional (3-D) memory organized in n memory modules (or memory slices), wherein each memory module has n 2 memory storage units as described in the above-identified references.
  • the cubic frame buffer also includes an independent dual access and addressing unit (not shown in the figures).
  • a 3-D skewed memory organization as described in the above-identified references, enables conflict-free access of any beam (i.e., a ray parallel to a main axis of the cubic frame buffer).
  • the apparatus also includes an inte ⁇ olation mechanism 28, shading mechanism 30 and projection mechanism 32.
  • the addressing unit of the cubic frame buffer 22 maps voxels in specific memory locations of the cubic frame buffer so as to provide conflict-free access of beams of voxels. Specifically, a voxel with space coordinates (x,y,z) is mapped onto the k- th memory module by:
  • the third coordinate guarantees that only one voxel from a corresponding beam resides in any one of the memory modules.
  • Each of the 2-D buffers 24 of the present invention are 2-D voxel storage devices having 2n 2 -n memory storage units.
  • the cubic frame buffer 22 is coupled to the 2-D buffers 24 by the interconnection mechanism 26.
  • the interconnection mechanism hereinafter referred to as the "fast bus" is an interconnection network that supports the high-bandwidth transfer of data (beams of voxels) from the cubic frame buffer to the 2- D buffer.
  • the fast bus manipulates the beams of voxels including skewing, de-skewing, fanning and de-fanning the voxel beams in order to support both parallel and perspective projection.
  • the fast bus employs multiplexers and transceivers with associated control units and a multi-channel bus to accomplish the data transfer speeds required for real-time volume visualization.
  • the voxels utilized in the volume visualization apparatus can be provided by a data acquisition device 23 (such as a scanner or M.R.I, device) or other mechanisms as known in the art.
  • a data acquisition device 23 such as a scanner or M.R.I, device
  • other mechanisms as known in the art.
  • the 512 memory modules of the cubic frame buffer are divided into 32 groups having 16 memory modules in each group.
  • memory modules 0-15 transfer voxel information on channel 0 of the fast bus
  • memory modules 16-31 transfer voxel information on channel 1 of the fast bus and so on such that memory modules 496-511 transfer voxel information on channel 31 of the fast bus.
  • the fast bus includes a plurality of multiplexers 25 such that the voxel data from the memory modules (0-511) of the cubic frame buffer 22 are time-multiplexed onto a designated 16-bit fast bus channel for that group of memory modules.
  • Table 1 shows the memory module data time-multiplexed on the fast bus.
  • the signal multiplexing is achieved by utilizing a clock input with transceivers 27 associated with each memory module.
  • the 2-D buffers are divided into 32 groups wherein each group includes 16 memory modules.
  • voxel data from the 32 channels of the fast bus are provided into a respective memory module of the group of 16 memory modules.
  • the operation of the multiplexers 25 and transceivers 27 are controlled by a look-up table commonly referred to as the bus channel map.
  • the bus channel map is pre-computed based upon selected viewing parameters (i.e., viewing direction, etc.). A change in the selected viewing parameters requires a re-computation of the look-up table. However, the limited size of the look-up table allows the re-computation to occur without affecting the real-time volume visualization and processing provided by the system.
  • the apparatus of the present invention 20 also preferably includes an inte ⁇ olation mechanism 28 coupled to the 2-D buffers 24.
  • the inte ⁇ olation device is a tri-linear (TRILIN) inte ⁇ olation mechanism which receives information about continuous viewing rays that are cast, preferably from the selected viewing position, through the cubic frame buffer 24. At evenly spaced locations along each viewing ray, sample points are indicated.
  • the inte ⁇ olation mechanism performs a tri-linear inte ⁇ olation utilizing voxel data corresponding to pixel points proximate to each viewing ray sample point in order to determine inte ⁇ olated voxel values for the sample points. These inte ⁇ olated voxel values are used for providing a more accurate volume visualization image.
  • the apparatus of the present invention may also include a shading mechanism 30 coupled to the inte ⁇ olation mechanism 28.
  • the shading mechanism can be coupled directly to each of the 2-D buffers 24.
  • the shading mechanism receives viewing ray and sample point information from the inte ⁇ olation mechanism. Thereafter, the shading mechanism receives the voxel data values proximate each viewing ray sample point. More specifically, the shading mechanism receives discrete voxel rays on the immediate left, right, above and below as well as the values of the viewing ray sample points proximate the current sample point. Based on the voxel information provided, a gradient vector for each sample point is determined by taking the differences of all voxel values proximate each sample point to provide an indication of the direction and amount of change in the characteristics of the object.
  • Figures 5A and 5B illustrate two different gradient estimation schemes.
  • the most simple approach is the 6-neighborhood gradient.
  • This method uses the difference of neighboring sample values along the continuous ray, P(n,m+1)-P(n,m- 1) in the x direction and P(n+l,m+l)-P(n-l,m-l) in the y direction.
  • P(n,m+1)-P(n,m- 1) in the x direction
  • P(n+l,m+l)-P(n-l,m-l) in the y direction.
  • the aliasing problem is circumvented by performing an additional linear inte ⁇ olation.
  • the additional step includes resampling the neighborhood rays at positions that are orthogonal (black samples) to the current sample point.
  • This approach is commonly referred to as the 10-neighborhood gradient estimation, and it solves the problem of switching the major axis during object rotations.
  • Figure 5B the use of a 26-neighborhood gradient will be described.
  • the aliasing problem is circumvented by performing an additional linear inte ⁇ olation.
  • the additional step includes resampling the neighborhood rays at positions that are orthogonal (white samples) to the base plane.
  • a bi-linear inte ⁇ olation or a linear inte ⁇ olation has to be performed.
  • These approaches are commonly referred to as the 12-neighborhood and 8 -neighborhood gradient estimation, respectively.
  • Each of these schemes serve to solve the problem of switching the major axis during object rotations and provide gradient vectors that are orthogonal to the main axis of the data set.
  • each projection ray plane PRP
  • PRP projection ray plane
  • the shading mechanism preferably also includes a light vector lookup table.
  • a light vector lookup table By knowing the gradient value and the values of the light vector look-up table, an intensity of each sample point can be generated using a variety of shading methods (e.g., using an integrated Phong Shader as known in the art).
  • opacity values are generated using a transfer function represented as a 2-D lookup table indexed by sample density.
  • the present invention also preferably includes a projection mechanism (RPC) 32.
  • the projection mechanism receives inte ⁇ olated voxel values (corresponding to viewing ray sample points) from the inte ⁇ olation mechanism 28, combines the inte ⁇ olated voxel values in one of the variety of ways, and generates a pixel value for each viewing ray.
  • the pixel value corresponds to the color, opacity and texture of the object or space being represented at a corresponding pixel location.
  • the projection mechanism is able to combine the inte ⁇ olated voxel values with either back-to-front compositing, front-to-back compositing, first opaque projection, weighted sum projection, last-to-first cut projection or first-to-last cut projection (which provides a volume visualization of a cross section of an object or region having a specified thickness).
  • the projection mechanism is a ray projection cone (RPC) which generates one pixel value signal per clock cycle using a variety of projection schemes as described above.
  • the ray projection mechanism includes a plurality of input ports 34 for receiving the plurality of inte ⁇ olated sample signals for each of the plurality of continuous viewing rays.
  • Each voxel combination stage includes a plurality of voxel combination units (VCU) 38. As shown in Figure 6, each successive voxel combination stage includes fewer VCU than the previous voxel combination stage.
  • VCU voxel combination units
  • each VCU is preferably coupled to three input ports 34.
  • the RPC is designed to select only two inte ⁇ olated sample signals from two of the three input ports or preceding VCU's. Specifically each VCU selects as input the left and center or right and center connection for receiving input signals depending upon the selected viewing scheme.
  • the RPC is a folded (circular) cross-linked binary tree with n leaves, which can be dynamically mapped onto a tree with its leftmost leaf at any arbitrary end-node of the RPC.
  • This allows the processing of a viewing ray of voxels such that the first voxel of the ray can be provided to any input port of the RPC.
  • This in turn allows the cone to be hardwired to the outputs of the 2-D buffers which contain the voxels.
  • Such a configuration eliminates the need for a set of n, n-to-1 switching units or barrel shifters for de-skewing of the 2-D buffer data as was required by prior art ray projection mechanisms.
  • the leaves of the cone contain the TRILIN and the shading mechanism.
  • the RPC accepts as input a set of n inte ⁇ olated sample signals along the viewing ray and produces a pixel value signal for the pixel corresponding to the associated viewing ray.
  • the cone is a hierarchical pipeline of n-1 primitive computation nodes VCU. At any given time frame, the cone is processing log n viewing rays simultaneously in a pipelined fashion, producing a new pixel value signal corresponding for one pixel of the display every clock cycle.
  • each voxel may be pre-stored with every voxel or provided through a look-up table or a transfer function inside the shading mechanism at the leaves of the cone.
  • the VCU produces an output signal by performing one of the following operations:
  • W is the weighting factor and k is the cone level (i.e., the number of voxel combination stages). W k is pre-computed and pre-loaded into the VCU's. It should be mentioned that a weighted sum is useful for depth cueing, bright field, and x-ray projections. Compositing is determined by the following:
  • the pixel value is transmitted, for example, to a general pu ⁇ ose host computer or a pixel processor 42, where post-processing, such as post-shading, splat- ting, and 2-D transformation or wa ⁇ ing, is performed.
  • the apparatus of the present invention may also include a frame buffer 40, pixel processor 42 and display device 44 coupled to the projection mechanism.
  • the pixel value signal generated by the projection mechanism 32 is provided to the frame buffer 40 where each pixel value signal is stored, provided to the pixel processor 42 for 2-D transformation, filtering or wa ⁇ ing, and thereafter provided to a display device 44 for visual display.
  • the pixel processor 42 transforms the pixel value signal so that it can be properly displayed on the display device.
  • the inte ⁇ olation mechanism 28 of the present invention generates a voxel value signal at non-voxel locations by utilizing the eight surrounding voxels and inte ⁇ olating as follows:
  • P abc P 000 (l-a)(l-b)(l-c)+P 100 a(l-b)(l-c)+ P 010 (l-a)b(l-c)+P 001 (l-a)(l-b)c+
  • the relative 3-D coordinate of a corresponding sample point within the cubic frame buffer with respect to a corner voxel closest to the origin is (a,b,c).
  • Different optimizations aim at reducing the arithmetic complexity of this method, but the arbitrary memory access to fetch eight neighboring voxels for each sample point makes this one of the most time consuming operations during volume rendering.
  • the data access time can be greatly reduced.
  • four discrete rays are fetched from the buffer, two from each of the projection ray planes (PRP) above and below the current ray.
  • the projection ray planes are provided from the 2-D buffers.
  • neighboring rays reside in adjacent inte ⁇ olation modules, requiring only a local shift operation of one voxel unit between neighbors.
  • FIG. 7A An inte ⁇ olation method is shown in Figures 7 A and 7B. Open circles 300 represent discrete ray A and black circles 304 represent discrete ray B. Shaded boxes 302 represent missing voxels.
  • sample points 46 on the continuous ray 48 are inte ⁇ olated using bi-linear inte ⁇ olation between samples of proximate discrete rays 50 (white) and 52 (black).
  • the first sample point of the continuous ray 48 can be correctly inte ⁇ olated using four voxels from discrete rays 50, 52 since the four voxels form a rectangle (i.e., the rays do not make a discrete step to the left or right).
  • the discrete rays step to the left or right as is the case for the second and fourth samples along the continuous ray 48, the four adjacent voxels form a parallelogram, and a straightforward bi-linear inte ⁇ olation might not provide accurate voxel sample values. Therefore, the grey shaded square voxels 54,56 are required to yield a more accurate result. However, these voxels reside on discrete rays two units away from the continuous ray 48.
  • FIG 7B the problem of not utilizing the best voxel point to provide an inte ⁇ olated sample signal is shown for perspective projections. Since discrete rays diverge for perspective viewing, the correct neighboring voxels are not stored in the 2-D buffers. For example, only two voxels 58, 60 of discrete ray 62 contribute to the correct inte ⁇ olation of the third sample point 64 of the continuous ray 66. In the 3-D case, as many as six voxels may be missing in the immediate neighborhood of sample point 64 for perspective projections.
  • each 3-D coordinate along the continuous ray consists of relative weights for linear inte ⁇ olations along each axis in possible sheared voxel neighborhoods. These weights can be pre-computed and stored in templates.
  • FIG. 8A-8D the steps necessary for inte ⁇ olation in 3-D are shown for both parallel projection (Figs. 8A and 8C) and perspective projection (8B and 8D).
  • Open circles 300 represent discrete ray A
  • black circles 304 represent discrete ray B.
  • four linear inte ⁇ olations are performed in the direction of the major axis (the Z-axis is the major direction of travel of the continuous ray) using eight voxels of four neighboring discrete rays stored the 2-D buffers.
  • these eight voxels are the vertices of an oblique parallelepiped for parallel projections or of a frustum of a pyramid for perspective projections.
  • voxels each reside on two separate planes one unit apart, which are commonly called the front 308 or the back 306 plane depending on when it is encountered during ray traversal in the direction of the major axis. Therefore, only one weight factor has to be stored, corresponding to the distance between the front plane and the position of the ray sample point.
  • the resulting four inte ⁇ olated values (double-hatched circles 312) form a rectangle and can be bi- linearly inte ⁇ olated to yield the final inte ⁇ olated sampled value 310.
  • This bi-linear inte ⁇ olation is divided into two linear inte ⁇ olations between the corner values (single- hatched circles 314) and the final linear inte ⁇ olation between the edge values.
  • this is shown as two inte ⁇ olations in the X-direction followed by one inte ⁇ olation in the Y-direction.
  • the sample points corresponding to the continuous rays are preferably inside the polyhedron defined by the voxels on the four surrounding discrete rays.
  • all continuous rays start at integer positions of the base plane (i.e., they coincide with voxels of the first slice of the volume dataset).
  • the use of these rays during ray casting effectively reduces the tri-linear inte ⁇ olation to a bi-linear inte ⁇ olation, because all sample points along the ray fall onto the front planes of the parallelepiped or pyramid frustum.
  • a continuous viewing vector 324 is split into a dx component along the X-axis (dx and dy in 3-d) and a unit vector in direction of the major axis (the Y-axis).
  • the viewing vector may be added to the current sample position in order to get the new sample position.
  • the addition of dx at the current sample position leads to a step of the discrete rays in the x direction. This step can only occur if the current sample position has a relative x offset with respect to the lower left corner voxel of more than 1 - dx for positive dx (or less than 1 + dx for negative dx).
  • dx and dy are the components of the viewing vector in the x and y directions, respectively. Notice that for 45 ° viewing angle, dx and dy are 1 , yielding an offset of 0 and bi-linear inte ⁇ olation as shown in Figure 9.
  • a single ray is cast from the origin of the image plane onto the base-plane using uniform distance between samples and the offset is chosen in the major direction of the first sample after the ray penetration of the base-plane. If necessary, the offset is iteratively reduced until it satisfies the above condition. This leads to view dependent offsets in the major direction of travel and to varying resampling of the dataset.
  • the variation of resampling points according to the viewing direction is an advantage for interactive visualization, because more of the internal data structure can be revealed.
  • Each discrete ray consists of n voxels, independent of the viewing direction. Since the maximum viewing angle difference with the major axis is not more than 45 degrees, the volume sample rate is defined by the diagonal through the cube and is by a factor of ⁇ / " 3 higher for orthographic viewing. It has been found that for ray-compositing, this is not an important consideration due to the averaging nature of the compositing operator.
  • the eight voxels surrounding the sample point either form a cube with sides of length one or an oblique parallelepiped as shown in Figure 8 A.
  • the surrounding voxels may form the frustum of a pyramid with parallel front and back planes as in Figure 8B. Due to the divergence of rays towards the back of the data set, the column spanned by this frustum increases, thereby reducing the precision of the tri-linear inte ⁇ olation.
  • the center of projection (C) and the field-of view (FOV) in perspective projections also influence the sampling rate.
  • the discrete line algorithm casts exactly one ray per pixel of the base-plane, or a maximum of 2n rays per scanline.
  • the discrete line algorithm in cases where the FOV extends across the dataset (correct sampling case 328), this guarantees better sampling than regular image order ray-casting which would cast n rays scanning the FOV and send wasteful rays that miss the dataset.
  • the discrete line stepping yields undersampling (case 330) in the active regions of the base-plane.
  • Case 332 of Figure 11 shows a situation where two base-plane images contribute to the final view image. This is the worst case in the generation of three base-plane projections for a single perspective image.
  • the apparatus and method of the present invention provides a more accurate representation of the object or scene being displayed due to the inte ⁇ olation and shading included in the invention.
  • the apparatus of the present invention since the present invention does not require conveyors or the like as used in prior art devices, the apparatus of the present invention operates more efficiently and faster than the prior art systems because the data manipulation is performed "on the fly" in a highly parallel manner. Specifically, the apparatus and method performs in O(n 2 ) time as compared to the prior art system which performs in O(n 2 logn) time.
  • the interconnection mechanism is capable of performing both de-skewing for parallel projection and de-fanning for perspective projection (i.e., a form of data compression)
  • the present invention is capable of supporting perspective projection and real-time visualization of four dimensional (4-D) data.
  • the apparatus of the alternative embodiment of the present invention preferably includes a cubic frame buffer (CFB) 22 having a plurality of memory storage units capable of storing voxel data therein as previously described.
  • the cubic frame buffer of the alternative form of the present invention may include a 3-D skewed memory organization enabling conflict free access of any beam of voxels 72 (i.e., a ray parallel to a main axis of the cubic frame buffer) as shown in Figure 13.
  • the apparatus also includes a first two-dimensional buffer 73, an inte ⁇ olation mechanism (TRILIN) 28, two two-dimensional buffers 24, shading mechanism 30, and compositing mechanism 74.
  • TRILIN inte ⁇ olation mechanism
  • Coupled to the output of the compositing mechanism is a pixel processor 42, frame buffer 40, and display device 44 for generating the three- dimensional (3-D) volume projection image.
  • the relative positioning of the pixel processor 42 and frame buffer 40 can be interchanged as required.
  • the method includes generating a viewing ray for each of the plurality of pixels of the display device 44. As each viewing ray traverses through the CFB, the viewing ray may change any of its x, y and z coordinates.
  • the plurality of viewing rays define a viewing direction through the cubic frame buffer.
  • the method further includes generating a plurality of regularly spaced sample points along each viewing ray. In order to assign a value signal to each sample point of the viewing ray, an inte ⁇ olation process is employed utilizing values assigned to voxels within the CFB. Specifically, beams of voxels are retrieved from the cubic frame buffer.
  • a beam of voxels is any discrete row, column or axle of voxels parallel to a primary axis (x, y or z) of the cubic frame buffer.
  • two n x 1 slices of voxel data signals 76 which include n beams of voxels (as shown in Fig. 13) are consecutively retrieved and provided to the inte ⁇ olation mechanism 28.
  • These slices of voxel data signals can be substantially parallel to the major axis of the 3- D memory (i.e., parallel to a normal of the base plane) or substantially orthogonal to the major axis of the 3-D memory (i.e., parallel to the base plane). While two entire n x 1 slices of voxel data are provided to the inte ⁇ olation mechanism and utilized for each of the plurality of viewing rays as described above, for simplicity the description will be limited to one group of viewing rays.
  • one n x 1 slice of voxel data signals 76 which includes n beams of voxels (as shown in Fig. 13) are consecutively retrieved and provided to the inte ⁇ olation mechanism 28.
  • These slices of voxel data signals can be substantially parallel to the major axis of the 3-D memory (i.e., parallel to a normal of the base plane) or substantially orthogonal to the major axis of the 3-D memory (i.e., parallel to the base plane).
  • the inte ⁇ olation mechanism is a bilinear or higher order device.
  • the inte ⁇ olation mechanism is a linear or higher order device which utilizes a larger neighborhood of voxels inside the n x 1 slice. It is foreseen that not only one slice of voxel data could be utilized, but that a plurality of slices can be accessed to generate the inte ⁇ olated sample point signals.
  • Each of the plurality of viewing rays 78 traversed through the cubic frame buffer 22 (and the corresponding sample points thereon) may be defined by two projection ray planes (PRP) within the CFB.
  • the two PRP are commonly referred to as a top PRP 80 and a bottom PRP 82.
  • the method of the present invention includes accessing a first beam of voxel signals 81 from the top PRP 80 and a first beam of voxel signals 83 from the bottom PRP 82.
  • Each beam of voxels 81, 83 is provided to a "back-face" input of the inte ⁇ olation mechanism as shown in Figure 14.
  • n is the size of the cubic memory.
  • the numbers in the planes indicate the x coordinate of a voxel that will be read from module k given certain y and z coordinates.
  • voxels from adjacent beams in the same PRP and voxels from adjacent beams from different PRPs have a different order of x coordinates (i.e, a skewing difference of one).
  • this skewing difference between adjacent voxel beams is compensated such that appropriate voxel values of each voxel beam (i.e., voxels having the same x coordinate value) are aligned by providing a shift when voxel data signals are accessed.
  • the alignment of voxel data signals is preferably achieved by providing the bottom PRP 82 to the back face of the inte ⁇ olation mechanism 28 without shifting the voxel data signal, and by providing the top PRP 80 to the back face of the inte ⁇ olation mechanism with a shift of one unit in the negative k direction.
  • the method of the present invention further includes moving the first beam of voxels of the first and second voxel planes from the "back face” to a "front face" of the inte ⁇ olation mechanism during the next clock cycle.
  • a second beam of voxels from the top PRP 80 and a second beam of voxels from the bottom PRP 82 are provided to the "back face" input of the inte ⁇ olation mechanism 28.
  • the first voxel beam of both the top and bottom PRP 80, 82 are shifted one position in the positive k direction when they are provided from the "back face” to the "front face” of the inte ⁇ olation mechanism. Therefore, when the four corresponding voxel beams are present in the inte ⁇ olation mechanism, the voxels of each voxel beam will be aligned so that the correct 8-voxel neighborhood is present to perform tri-linear inte ⁇ olation for a sample point of each viewing ray. As a result, four voxel beams simultaneously reside in the inte ⁇ olation mechanism.
  • an inte ⁇ olated sample point signal for the first sample point of each viewing ray is generated by the inte ⁇ olation mechanism 28.
  • the inte ⁇ olation mechanism 28 is a tri-linear inte ⁇ olation mechanism (TRILIN).
  • TRILIN tri-linear inte ⁇ olation mechanism
  • a bi-linear inte ⁇ olation or higher order inte ⁇ olation mechanism using voxels that are inside or outside the eight voxel neighborhood i.e., less than or more than 8 voxel values
  • a simple stepping scheme is employed to properly align the voxel beams in adjacent voxel planes.
  • the stepping scheme from the cubic frame buffer to the "back-face” input of the inte ⁇ olation mechanism and from the "back-face” input to the "front-face” of the inte ⁇ olation is as follows:
  • the method further includes generating a gradient estimation signal for each inte ⁇ olated sample point signal by providing the inte ⁇ olated sample point signals to a buffer 84 for temporary storage therein.
  • the buffer preferably includes an above buffer 86, current buffer 88 and below buffer 90. However, the method can be implemented with any two of these three buffers.
  • the gradient estimation signal provides an indication as to surface inclination at a specific inte ⁇ olated sample point. In a preferred form of the present invention, three pairs of inte ⁇ olated sample point signals about a specific inte ⁇ olated sample point are required to generate a gradient estimation signal.
  • the actual differences in all directions can be computed by determining central differences between selected inte ⁇ olated sample points.
  • the shading mechanism In order to determine the gradient difference in the major direction (i.e., along the z viewing direction), it is necessary for the shading mechanism to have access to two beams of inte ⁇ olated sample points within the current buffer that are two clock cycles apart (i.e., one beam that precedes and one beam that lags the desired sample).
  • Figure 14 illustrates that one group of inte ⁇ olated sample points is simultaneously output by each of the above, current and below buffers 86, 88, 90 to the shading mechanism 30.
  • the inte ⁇ olated sample point signals are not aligned in the above, current and below buffers as required.
  • the inte ⁇ olated sample point signals of the current buffer are preferably shifted one position in the positive k direction within the shading mechanism 30.
  • the below buffer sample point signals are preferably shifted two positions in the positive k direction within the shading mechanism.
  • the inte ⁇ olated sample point signals from the above and below buffers are preferably delayed by one clock cycle so as to be aligned with the inte ⁇ olated sample point signals in the current buffer.
  • the combination of the shifts and delays compensates for the skewing difference between the inte ⁇ olated sample point signals that are provided by the above, below and current buffers.
  • the shading mechanism preferably also includes a light vector lookup table.
  • an intensity of each sample point can be generated using a variety of shading methods (e.g., using an integrated Phong Shader as known in the art).
  • opacity values are generated using a transfer function represented preferably as a 2-D or 3-D lookup table or other indexing method to accessed by sample density and/or gradient value.
  • a simple stepping or shifting scheme is employed to properly align the inte ⁇ olated sample value signals with adjacent groups of inte ⁇ olated sample value signals in the shading unit.
  • a shift of inte ⁇ olated sample value signals is preferred from the above, below and current buffers to the shading mechanism 30 is as follows:
  • each of the inte ⁇ olated sample point signals including a shading component is provided to a compositing mechanism 74 which performs composition operations of sample points along respective viewing rays. Since the beams of shaded sample value signals that are output by the shading mechanism are still skewed, the composition of a single orthogonal ray cannot be done by a single compositing unit. Specifically, in a preferred embodiment of the invention, each inte ⁇ olated sample point of the viewing ray is composited in a different compositing mechanism.
  • composition of inte ⁇ olated sample points performed within the composition mechanism occurs in substantially the same method as previously explained.
  • the composition mechanism could combine the inte ⁇ olated voxel values with either back-to-front compositing, front-to-back compositing, first opaque projection, weighted sum projection, last-to-first cut projection or first-to-last cut projection or any other known composition technique.
  • inte ⁇ olated sample points are composited, they are provided, for example, to a general pu ⁇ ose host computer or a pixel processor 42, where post processing, such as post-shading, splatting, and 2-D transformation or wa ⁇ ing, is preformed.
  • post processing such as post-shading, splatting, and 2-D transformation or wa ⁇ ing
  • Pixel processing for display on the display device 44 preferably occurs as previously explained.
  • one slice of voxel data signals are accessed from the 3-D memory.
  • Each slice of voxel data signals is substantially parallel to the base plane of the 3-D memory.
  • the slice of voxel data consists of a plurality of voxel data beams that are provided to the inte ⁇ olation mechanism.
  • the inte ⁇ olation mechanism has only one face (either a back face or a front face).
  • the inte ⁇ olation mechanism is a bi-linear or higher order device for parallel viewing wherein the inte ⁇ olation utilizes voxels from one slice.
  • the inte ⁇ olation mechanism is a linear or higher order device which utilizes a larger neighborhood of voxels from the one slice.
  • the inte ⁇ olated voxel value signals are thereafter provided to the above, below and current buffers 86, 88, 90 as previously described.
  • the above described method can support both parallel and perspective viewing of objects stored in the 3-D memory.
  • the slices of voxel data signals accessed for any one viewing ray may not be utilized for determining the inte ⁇ olated sample valve signal for any other viewing ray. Therefore, each viewing ray requires access of individual voxel planes. This is also required for the shading mechanism and compositing mechanism.
  • Figure 16 shows the implementation of five memory units 75 of the CFB 22 including a first in - first out buffer (FIFO buffer) 84.
  • FIFO buffer first in - first out buffer
  • Each of the five memory units 75 are unable to output two beams simultaneously but by including the FIFO buffer, one slice of memory is provided for one of the PRPs.
  • the apparatus also includes five inte ⁇ olation mechanisms (TRILIN) 28 that are coupled to the FIFO buffers and memory units in accordance with the shifting requirements between the cubic frame buffer and the inte ⁇ olation mechanisms as previously described.
  • TRILIN inte ⁇ olation mechanisms
  • the five inte ⁇ olation mechanism 28 are coupled to five shading mechanisms 30 utilizing direct connections for one of the above, below and current planes, and two FIFO buffers 86 for the two remaining above, below and current planes.
  • the shading mechanisms 30 are interconnected among proximate shading mechanism in accordance with the shifting requirements between the TRILIN and the shading mechanism as previously described.
  • the apparatus also includes five compositing units 74 that are preferably directly coupled to a respective shading mechanism for receiving an output signal from each shading mechanism. Each compositing unit perform a projection method as previously described. The interconnection of each component is evident based upon the above requirements of data shifts by and between each component of the apparatus.
  • the output signals provided by the compositing mechanism are output to a pixel bus 88 that forwards the compositing mechanism output signals to the pixel processor 42, frame buffer 40 and display device 44 to generate the volumetric image.
  • the alternative embodiment of the present invention is advantageous because it is a novel ray-casting method and apparatus that leads to a uniform mapping of sample points along viewing rays onto voxels that avoids multiple access of voxel data for the inte ⁇ olation mechanism and shading mechanism.
  • the method and apparatus of the present invention access each voxel inside the cubic frame buffer (CFB) only once per projection. This permits regular memory access patterns leading to high efficiency and enabling real-time performance rates needed for visualization of dynamically changing 3D data sets.
  • Ray-slice-sweeping is a plane sweep algorithm for volume rendering.
  • the compositing buffer sweeps through the volume and combines the accumulated image with a new slice of just-projected voxels.
  • the image combination is guided by sight rays from the view point through every voxel of the new slice. All voxels of the data set contribute to the rendering.
  • Gradients are computed at the voxel positions, thereby improving accuracy and allowing a more compact implementation than previous methods. Less control overhead is required than previous methods.
  • Ray-slice-sweeping is a hybrid method in which each volume slice projects all its voxels towards the view point, but only one inter-slice unit in distance.
  • the compositing buffer contains the base plane image which then has to be wa ⁇ ed onto the image plane.
  • the sweep is done in object storage order, yet it is image- pixel driven, while the wa ⁇ is performed in scan line order; thus, ray-slice-sweeping is properly described as a hybrid method.
  • the ray-slice-sweeping method has particular utility for perspective projections. However, parallel projections can also readily be performed with the view point at infinity.
  • FIG. 17 depicts ray-slice-sweeping through a volume using front-to-back compositing.
  • Case 400 is for a parallel three dimensional projection
  • Case 402 is for a perspective three dimensional projection
  • a three dimensional volume 404 includes individual voxels located on a grid 406.
  • Figure 17 depicts the three dimensional volume in a two dimensional view pe ⁇ endicular to the y axis, showing the x and z directions.
  • Each volume 404 has a base plane 408 employed for projection pu ⁇ oses to be discussed below.
  • a plurality of sight rays 410 are cast from a view point 412 for each slice of voxels.
  • the view point in Case 400 of Figure 17 is at infinity, since this is a parallel projection.
  • Slices of voxels are indicated as reference numeral 414. While sweeping through the volume 404 in, for example, front-to-back processing order (i.e., front-to-back sweeping), each voxel of the current slice has to be colored, illuminated and classified, resulting in RGB A intensities for that voxel (red, green, blue and alpha). In a preferred form of the method, sweeping through the dataset is carried out in the positive z direction, independent of the view point.
  • front-to-back sweeping and front-to-back compositing are used to combine the RGB A values of the current slice with those of the compositing buffer.
  • Front-to-back compositing is shown in greater detail in Figure 19 and will be discussed below.
  • Figure 18 shows back-to-front compositing with the view point 412 in back of the three dimensional data set 404. Similar items have received similar numbers.
  • the parallel case is indicated as 416 with the perspective case indicated as 418.
  • rays 410 cast by the method within a horizontal cut through the volumetric dataset 404 are depicted.
  • the data slices 414 are preferably processed in the positive z direction. Those sight rays 410 which are dashed exceed a 45° viewing angle limit, and do not influence the final image.
  • the sight rays 410 point towards the voxel grid positions 406 and start in between voxels of the previous slice.
  • an inte ⁇ olation such as a bilinear inte ⁇ olation
  • compositing will be discussed below with respect to Figure 19. It will be appreciated that all values written to the compositing buffer are aligned with the voxel grid 406. Note that the rays depicted in Figures 17 and 18 lie along the sight rays 410 but extend only between adjacent slices 414. These short portions of the sight rays are referred to herein as interslice ray segments to avoid confusion, and will be discussed further below.
  • sight rays 410 are depicted for a view point 412 behind the volumetric dataset 404.
  • the sight rays begin at the voxel positions and point towards the previously processed slice.
  • the accumulated colors employed in back-to-front compositing are determined, for example, by bilinear inte ⁇ olation of the four compositing buffer colors surrounding the endpoint of a given ray 410.
  • FIG. 19 depicts both front-to-back and back-to-front compositing techniques suitable for use with the method of the present invention.
  • Case 424 depicts front-to-back compositing
  • Case 426 depicts back-to- front compositing.
  • the interslice ray segment 428 can extend along a sight ray cast from the view point 412 (not shown in Figure 19) through an intersection point 430 in the compositing buffer 432 to a grid point location 434 of the current slice 436.
  • inte ⁇ olation can occur in the compositing buffer using the formulas depicted in the figure.
  • the interslice ray segment 428 can extend along a sight ray cast from the view point 412 (not shown in Figure 19) through a grid point location 434 in the current slice terminating in an intersection point 430 within the compositing buffer, wherein inte ⁇ olation is again performed, using the equations shown in Figure 19. It is to be understood that higher or lower orders of inte ⁇ olation may be performed, as discussed further below.
  • a discussion of the architecture of the apparatus of the present invention will now be presented. The invention disclosed in the parent of the present application, i.e., U.S. Patent Application Serial No.
  • 08/388,066 filed February 13, 1995 is a pipelined scalable volume rendering architecture based on ray casting.
  • the present invention modifies the ray-casting approach by implementing ray-slice-sweeping. Sample points are employed which are directly on the voxel grid. Thus, the trilinear inte ⁇ olation stage (see items 28 in Figure 16) is not needed in the apparatus of the present invention. In turn, the gradient computation is simplified. During perspective projection, all voxels of the dataset contribute to the rendering. Since there is less control information required for rays, the implementation can be more compact.
  • Voxel memory is distributed over several memory modules 438.
  • a rendering pipeline 440 is provided for each memory module 438.
  • each rendering pipeline 440 is preferably implemented on a separate chip.
  • Each of the rendering pipelines 440 works on the three dimensional projection simultaneously. Only nearest neighbor connections are required for horizontal communication, until final base plane pixels are computed. These pixels are sent over a global pixel bus 442 to a host or graphics card memory 444 to be assembled into one image and wa ⁇ ed onto the final image plane as depicted in Figures 17 and 18.
  • signal flow graphs As are known in the art of volume imaging, will be employed.
  • SFG signal flow graphs
  • circles represent stages within a pipeline 440
  • vertical arcs show data flow within a pipeline
  • diagonal arcs show data flow between pipelines.
  • the weight on each arc represents the number of pipeline cycles for which the data must be delayed before reaching the next pipeline stage. All units assume partial beam processing.
  • Beams are rows of voxels. Breaking a beam into equal size segments produces partial beams; the size of a partial beam is equal to the number of parallel pipelines 440 implemented in hardware.
  • FIG. 21 depicts partial beams 446 having p voxels and full beams 448 having n voxels.
  • slices 414 are preferably incremented in the positive z direction.
  • full beams are preferably incremented in the positive y direction.
  • partial beams 446 are preferably incremented in the positive x direction.
  • each module has an extension unit (EX) at every pipeline stage with neighbor connections.
  • partial-beam-end refers respectively to the end of a partial beam, the end of a full beam and the end of a slice.
  • Each voxel has control bits associated with it; when the end of a partial beam is reached, the partial- beam-end control bit is set to true; when the end of a full beam is reached, the beam-end control bit is set to true and when the end of a slice is reached, the slice-end control bit is set to true.
  • EX extension unit
  • a full beam generally includes all the voxels in one row, while a partial beam is the result of dividing a full beam into a number of smaller, partial beams; these may be equal in size or not.
  • beam refers to a full beam or the general case of either a full or partial beam.
  • beam is intended to include at least one of a full beam and a partial beam. Note that any details about the architecture of the present invention which are not specifically discussed are identical to those of the architecture for the parent 08/388,066 case, with specific reference to Figures 12-16 as discussed above.
  • Figure 22 also depicts a three dimensional cubic frame buffer 450 which contains the individual memory modules 438, each of which connects to one of the rendering pipelines 440.
  • Cubic frame buffer 450 contains the memory modules 438 which provide distributed skewed storage of voxels, as well as an address generator 452 and a control unit 454.
  • the cubic frame buffer 450 is the only stage within the rendering pipeline (in this context, used to include the entire architecture) which has global position information. It makes all global decisions for all pipeline modules. Subsequent stages make their selections based on the control bits generated in the cubic frame buffer.
  • N s current slice number
  • N B current beam number
  • N b current partial beam number.
  • the view point V is also defined in the pipeline coordinate system.
  • the sight ray S of a voxel is the vector from the view point to the voxel position.
  • Voxel/sample position 456 is defined by the vector P in the pipeline coordinate system 460.
  • View point 412 is defined by the vector V in the pipeline coordinate system 460.
  • These normalized x and y components determine bilinear (or other order) inte ⁇ olation weights generated by control unit 454 (in Figure 22) and are forwarded to the compositing unit 464 (also in Figure 22).
  • the components dx and dy are also range checked. If the absolute value
  • At least one face of the three dimensional data cube 404 has a normal within plus or minus 45 ° of the sight ray S 458.
  • the slices 414 processed by the method of the present invention are parallel to that slice (i.e., that face of the dataset 404). Accordingly, the restriction to using only nearest neighbors is feasible.
  • all sight rays S 458 are the same; sweeping the volume 404 once delivers a complete image.
  • the sight rays S 458 differ for each voxel within the volumetric dataset 404, as might those faces of the volume 404 for which the normal is within plus or minus 45°.
  • each rendering pipeline 440 has only one dedicated memory module 438.
  • the weight and control portion 454 must also set the invalid bit for all voxels that have a direct neighbor on the other end of the dataset.
  • the cubic frame buffer 450 may also have to set the SR or ER bits. That is, ER should be set to true if the current voxel is on the left most slice of the dataset 404 and dx less than 0.
  • the SR and ER bits are forwarded to the compositing unit 464, where they enable the compositing buffer reset (this is done by SR) and the pixel output to the final image (this is done by ER).
  • the control unit 454 also has to set the start of beam (BS) bit whenever the current partial beam 446 is the first on a full beam 448. This bit is needed in the aforementioned extension units.
  • BS start of beam
  • the cubic frame buffer 450 must compute the pixel address for each compositing buffer element.
  • the coordinates are:
  • each pipeline 440 also preferably includes an ABC gradient estimation unit 468, a classification unit 470, an inte ⁇ olation unit 472 and a compositing buffer 474.
  • the classification unit 470 can include a look-up table for opacity based on intensity. It can, if desired, be formed as part of the shading unit 466.
  • the gradient estimation unit 468 can also be included as part of the shading unit 466, if desired. In this case, the shading unit deals with RGB, ⁇ , and other displayable parameters. Further details regarding the function of the gradient estimation unit 468 will now be provided. Referring to Figure 24, the preferred gradients used for shading in the present invention are central difference gradients.
  • the voxels needed for the gradient computation reside in three planes, the ahead 464, behind 466 and current 436 planes (slices).
  • the terminology ahead, behind and current reflects the processing order of the slices.
  • the slices may be stored in first-in- first-out (FIFO) buffers designated as ABC for ahead, behind and current respectively. Most preferably, only the behind and current slices are stored in a FIFO; the ahead slice preferably comes directly from the cubic frame buffer 450.
  • FIFO buffers will be discussed further hereinbelow with reference to Figure 46.
  • FIG. 25 depicts the spatial and temporal relationships between the gradient computation parameters with skewing considered.
  • SFGs signal flow graphs
  • Figure 27 the approach of Figure 26 is modified so that only nearest neighbor connections are used.
  • the approach of Figure 27 requires one additional pipeline stage to receive voxel d from the pipeline to the right and send it to the pipeline to the left.
  • the forwarding directions are changed such that the y-gradient is computed in that pipeline which also holds the corresponding center voxel m.
  • FIG 29 illustrates two full beams 448 in skewed space, each having five partial beams 446.
  • the shaded circles represent voxels with different intensities.
  • each partial beam 446 has four voxels.
  • the arrows indicate that a given voxel is the left neighbor of the other given voxel to which the arrow is connected.
  • the voxel in the right most pipeline is usually the left neighbor of the voxel in the left most pipeline one clock cycle later. Consequently, forwarding to the right requires buffering the right most voxel for one clock cycle. This buffer logically extends the next partial beam.
  • Figure 31 depicts a SFG employing the ideas described for wrap-around connections between the ends of the partial beams, just illustrated with respect to Figure 30.
  • Figure 31 also depicts the above-described extension unit EX.
  • the extension unit is designated by reference character 472.
  • extension unit 472 has a first input 474 for receiving the input data x, and a second input port 476 for receiving the BS (or other) bit from the control unit 454. Normally, input data x is forwarded directly to first output port 478.
  • the extension unit EX 472 functions similarly for any of the control bits employed in the present invention, for example, the partial-beam-end, beam-end or slice-end control bits discussed above.
  • Figure 32 depicts a first alternative for complete x, y and z gradient computation in a single SFG.
  • shading and classification apparatus and method according to the present invention will now be discussed. Note that shading and classification are identical to the shading and classification techniques set forth hereinabove with respect to the parent
  • the shading stage or unit 466 and its inter-relationship to other components in the pipeline 440 are depicted in Figure 22 above.
  • the shader 466 uses the intensity as an index into RGB color tables, and the gradient as an index into a reflectance map.
  • the color tables map different intensities to different colors.
  • the color transfer function can be used to segment the data on the fly.
  • the reflectance map is a quantized solution table for the illumination equation for rays along the main viewing direction and any surface orientation.
  • the final color of a shaded voxel is then composed from the values returned from these tables.
  • the most compact implementation just multiplies the RGB values from the color table with the intensity taken from the reflectance map. Reference should also be had to the above discussion of the gradient, classification and shading units 468, 470, 466 respectively.
  • Pixels are typically represented by RGB A where A represents the opacity ⁇ .
  • the opacity for each voxel is determined by a classification function which depends on the voxel intensity and the gradient magnitude. This function is stored in another look-up table. Filling the table using different algorithms allows very flexible viewing modes such as X-ray, surface, fog and the like.
  • RGBT can instead be employed in the apparatus and method of the present invention, wherein T represents the transparency t, transparency t being equal to 1- ⁇ .
  • Transparency can, if desired, be substituted for opacity ⁇ inasmuch as it simplifies compositing calculations.
  • the transparencies for each sample can be taken from a "transfer function" look-up table which is indexed by the sample value and the gradient magnitude at the given position. Such a transparency table also permits the flexible viewing modes discussed above for the opacity table.
  • the compositing employed with the present invention will now be discussed. Either back-to-front or front-to-back compositing can be employed. The choice will normally depend on the major viewing direction. If, in the dataset coordinate system 462 the view point 412 is left, above or in front of the dataset 404, front-to-back compositing is normally used. When the view point 412 is located other than those locations, back-to-front compositing is normally employed.
  • the view point when, in the dataset coordinate system, the view point is left, above or in front of the dataset 404, the view point will be referred to as being in front of the three dimensional buffer 450 and when the view point is right, below or behind the dataset 404, the view point will be referred to as being behind or in back of the three dimensional buffer 450.
  • the preferred compositing directions assume front-to-back sweeping through the slices 414 of the three dimensional volumetric dataset 404.
  • back-to-front sweeping back-to-front compositing can be used with the view point 412 in front
  • front-to-back compositing can be used with the view point 412 in back of the dataset 404.
  • back-to-front compositing the following equations are employed:
  • A (l-A acc )A new +A acc .
  • C new is used for any of the shaded voxel color components (RGB), and the term A new is used for the classified voxel opacity.
  • the subscript "ace” is used for the corresponding accumulated values taken from the compositing buffer 474.
  • the box 484 labeled bilinear inte ⁇ olation is intended to include the inte ⁇ olation units 472 for each pipeline 440. It is to be understood that other types of inte ⁇ olation, besides bilinear inte ⁇ olation, can be employed.
  • the circles containing S k indicate the new pixel from the shader and classification stages 466, 470.
  • the circles containing C k indicate composition of new and accumulated pixels.
  • the preferred form of compositing buffer 474 is a FIFO with the capacity for a full slice of RGB A values.
  • the compositing unit 464 preferably writes new values to the bottom of the slice FIFO compositing buffer 474, while the inte ⁇ olation unit 472 preferably reads from the top.
  • the inte ⁇ olation unit 472 of the present invention will now be discussed. It is preferred that the inte ⁇ olation unit 472 be a bilinear inte ⁇ olation unit.
  • the aim of the bilinear inte ⁇ olation unit is to determine the color at the intersection between the compositing buffer plane 432 and the sight ray through the already shaded voxel (refer to Figure 19). It should be recalled that, inasmuch as the viewing angle is limited to + 45°, the sight ray increments dx and dy are bounded. The intersection can therefore only occur within a 3 X 3 region around the current voxel.
  • Figure 35 depicts the skewed relative positions of the current voxel m and voxels a-i, from the previous slice, which may possibly influence voxel m.
  • Figure 35 shows the skewed relationship of the voxels within the aforementioned 3 X 3 region.
  • a minimum of four pipeline stages are necessary; three which forward their data to their right neighbors and one which forwards its data to the left. This optimal configuration is depicted in Figure 37, to be discussed below.
  • Figures 36 and 37 it will be appreciated that the possible region can be reduced from a 3 X 3 region to a 2 X 2 region merely by evaluation of the sign of dx and dy.
  • Figures 36 and 37 The second two stages in those figures then compute the actual bilinear inte ⁇ olation from those four compositing buffer pixels.
  • Figures 19 and 36 present the appropriate equations for the dependency of the inte ⁇ olation weights from the sight ray increments.
  • Figure 36 depicts the functionality of the pipeline stages 440 of Figure 37 assuming back-to-front compositing. For front-to-back compositing, dx and dy have the opposite sign, slightly changing the computations.
  • FIG. 38 depicts a comparison of the energy distribution in the ray-slice-sweeping method 490 of the present invention, contrasted to ray-casting techniques 492.
  • each voxel can only contribute to a maximum of four rays. Accordingly, the voxel energy distribution on the image plane has very sha ⁇ edges.
  • the energy of a voxel is concentrated along the ray direction, but with each slice it spreads to the neighboring voxels. This spreading is caused by the bilinear (or other) inte ⁇ olation.
  • FIG. 39 depicts back-to-front processing of the data set 404.
  • a slice of partially composited rays is stored in each step of the method, and as the next slice of data 414 is processed, it is composited with the slice buffer.
  • the slice-order method of the present invention only requires each voxel 494 to be accessed once, it does require more buffer storage.
  • the data is not sampled in a regular fashion precisely along rays from the eyepoint through an image pixel and then through the three dimensional data 404.
  • Hardware architectures of the present invention suitable for use in practicing the method of the present invention, can have separate pipelines/processors for each grid coordinate, so that the compositing information is buffered onto that chip-discrete grid coordinate.
  • slice-order ray-casting Another preferred form of apparatus and method in accordance with the present invention, designated as slice-order ray-casting, will now be described.
  • the slice-order ray-casting method is a hybrid which combines data sampling in a manner similar to the previously-described ray casting of the parent and grandparent applications, with the voxel access pattern of slice-order processing just described above. This is accomplished by casting a ray through the data, but storing the information at discrete grid positions. Since the processing is still in a slice-order fashion, the data is sampled in each slice at the point where the ray would intersect the current slice plane by using bilinear (or other) inte ⁇ olation, instead of on the exact grid coordinates like other slice- order methods.
  • Figure 40 represents a constant y coordinate plane within the three dimensional volumetric data 404, for a case of z-access slice-order processing.
  • the method begins by creating an initial compositing buffer for slice 3.
  • the initial compositing buffer contains the shaded and classified points sampled on the grid coordinates and composited with the background.
  • a ray 496 is shown from the voxel at location X Q in slice 3 to the eyepoint (i.e., view point) 412.
  • the black dots represent the gridpoint coordinate in which the data for the ray 496 is stored. It will be seen that for the first two steps, corresponding to slices 2 and 3, the data is stored at the X 0 grid coordinate.
  • the data for the ray 496 is shifted to the grid coordinate X, as indicated by the black dot now contained at X, in slice 1.
  • the data shifts to the new storage position its offset is adjusted by the unit distance; in this case, the data was offset 0.6 units at the time when it was shifted, with 0.6-1 being equal to -0.4.
  • the data was offset 0.6 units at the time when it was shifted, with 0.6-1 being equal to -0.4.
  • parallel projection there is a regular data movement pattern among the rays.
  • Ray-splitting is the idea of adaptive supersampling in order to maintain a uniform ray density.
  • the ray In front-to-back compositing, the ray is split into two child rays when neighboring rays diverge beyond some threshold.
  • the reference "An Efficient Method for Volume Rendering Using Perspective Projection" by K.L. Novins, F.X. Sillion and DP. Greenberg, of the Program of Computer Graphics at Cornell University in Ithaca, New York 14850, published in Computer Graphics. SIG- GRAPH'90 pages 95-100 sets forth a method intended for use with software volume rendering systems in order to handle problems with main-memory storage.
  • the ideas set forth therein can be adapted to deal with the merging problem in the present invention.
  • splitting rays One problem with splitting rays is that it becomes difficult to prevent split rays from colliding with themselves in the future. In order to prevent this, a local method is not sufficient, but a global method is too complex for real-time interaction. Ray-merging is similar to ray-splitting, except that it runs in the front-to-back order. When two or more rays suddenly attempt to utilize a single grid coordinate storage location, the two parent rays are averaged, spawning a new child ray. The new ray with averaged intensity, opacity and offset is shot from the current grid location directly toward the eye. A further problem encountered with perspective projections is that sometimes a ray shifts into a neighboring grid coordinate, and there is no other ray shifting into its original grid coordinate.
  • FIG 41 As depicted therein, a large number of rays can merge in a perspective, merging-ray tree.
  • the processing order is from the top of the diagram to the bottom, getting closer to the eyepoint (i.e., view point) 312 with each step.
  • the rear slice 498 which is located at the top of the figure, is initialized by generating a plurality of new rays (true rays designated by dotted lines), with zero offset, at each grid position 434 (indicated by open circles).
  • the ray newly spawned at 510 merges with its left neighbor.
  • the left subtree represents the contribution of four voxels and the right subtree, one. If we simply took the average intensity of the two ray subtrees, one voxel contribution is over represented and the other four are under-represented in the merged ray.
  • a bilinear inte ⁇ olation can then be performed on the 4 pixels surrounding the intersection, weighted based on distance from the intersection point.
  • the open squares represent pixels of the base plane 408, while the x's represent voxels which are processed but do not contribute to the final image as they lie beyond 45 degrees.
  • the "x" voxels can contribute in a second rendering path with a different base plane, as in the parent 08/388,066 application.
  • a simulation has been developed to evaluate ray merging decisions for arbitrary eyepoints (i.e., view points). Using the simulation, various statistical properties of the method of the present invention were evaluated on a 256 voxel square two dimensional grid. It was noted that 10.1% of newly spawned rays were encountered; 13.3% two-way merging steps were encountered; and 6.4 x 10 "7 % of three-way merging steps were encountered. An average of 8.1 slices were traversed between merges. The percentages are based on the total number of voxel positions in the 3-D grid.
  • the ray-merging method depicted in Figure 41 also performs some number of bilinear (or other order) inte ⁇ olations, but significantly less often than every step.
  • the ray-merging method depicted in Figure 1 benefits from some inherent anti-aliasing On average, one merging step is performed every 8.1 slices. This has the benefit of leading to an anti-aliased image..
  • the eye i.e., view point
  • Slice n is designated as 514 and is the plane containing the current data.
  • Slice 516 contains the values associated with the compositing buffer.
  • Ray 518 is shot from view point 512 through gridpoint locations 520 in the current data plane slice n. It intersects the compositing buffer at the location marked with crosshairs; and it is clear that the RGB and ⁇ energy for datapoint 520 is spread out to the four adjacent grid coordinate points in the compositing buffer which are designated by the dotted lines from the crosshair.
  • the slice-order energy-splatting approach saves on voxel buffering and also produces more accurate gradient measurements.
  • the compositing buffer is slid forward to get ready for the processing of the next slice forward in the data.
  • a ray is shot toward the eyepoint. Offsets of where the data is stored at each grid coordinate position are again accumulated and stored.
  • the compositing buffer data slides along rays and stores itself in the closest grid coordinate to the current position of the ray. Merging and new rays are handled in the same way as in ray-casting.
  • a full three dimensional simulation of the method was developed and implemented on a Silicon Graphics work station.
  • the method performs volumetric ray- casting and is suitable for both software and hardware utilizing a fixed number of rays in a slice-by-slice processing order.
  • a constant number of rays per slice produces perfect load balancing when mapping the rays to a constant number of processors like many currently proposed parallel architectures.
  • This method minimizes storage requirements because the preferred bilinear inte ⁇ olation requires less buffering than trilinear inte ⁇ olation. Further, bandwidth requirements are reduced because each voxel is read only once.
  • the method is independent of the classification and illumination model, and the resulting image is a base plane image of size n 2 , which is then projected onto the image plane by a two dimensional wa ⁇ .
  • the method and apparatus of the present invention previously described above, will now be summarized with reference to Figure 43 and the previously-discussed figures.
  • the ray-slice-sweeping method of the present invention generates a three- dimensional volume projection image having a plurality of pixels.
  • the image is generated from a view point 412, and utilizes discrete voxels 494, arranged in a three dimensional volume 404 which can be stored in the previously-discussed cubic frame buffer 450.
  • Each of the voxels 494 has a location and at least one voxel value associated with it.
  • the first step of the method includes selecting viewing and processing parameters which define the viewpoint 412, at least one base plane 408 of the three dimensional buffer containing the volume 404, with the base plane 408 being employed for projection pu ⁇ oses; and first and last processing slices 522, 524 of the three dimensional buffer 450.
  • first slice 522 would normally be considered as slice zero
  • last processing slice 524 would normally be considered as slice n-1.
  • n is equal to 4.
  • a further step of the method includes initializing a compositing buffer, such as compositing buffer 474 discussed above, which has a plurality of pixels, each of which has at least color, transparency and position associated with it.
  • a further method step includes sampling voxel values from three dimensional buffer 450 onto a current slice of sample points parallel to the first processing slice in order to provide sample point values.
  • slice 1 reference character 5266 is the current slice, parallel to the first processing slice 522, onto which the voxel values are sampled from the three dimensional buffer 450.
  • Figure 43 depicts a case where slice 1, reference character 526, is the current processing slice. It is to be understood that during the first execution of the sampling step, the first processing slice (slice 0, reference character 522) would be the current slice.
  • initializing could include, for example, combining with the background.
  • a further step in the present method includes combining the sample point values from the sampling step with pixels of the compositing buffer along a plurality of interslice ray segments 528.
  • Interslice ray segments 528 extend only between the current slice 526 and an adjacent slice, here, 522, which is associated with the c- ompositing buffer.
  • slice 0 reference character 522
  • the first processing slice would have been initialized into the compositing buffer, preferably by combination with the background.
  • each voxel 494 in the current slice would normally have an associated interslice ray segment 528 which would extend to the compositing buffer slice (here slice 0, reference character 522). Only three interslice ray segments 528 have been shown, for clarity.
  • the sampling step includes sampling voxel values from the three dimensional buffer 450
  • the combining step preferably includes the substep of casting a plurality of sight rays 528 from the view point 412 through the grid points in the current slice of sample points 526.
  • the interslice ray segments 528 then extend along the sight rays 530 from the sample points of the current slice 526 to the adjacent slice 522 associated with the compositing buffer.
  • the interslice ray segments 528 intersect the adjacent slice 522 at a plurality of intersection points 532, designated by crosses. Only three intersection points 532 are shown in Figure 43, for clarity.
  • the step of repeating the sampling and combining steps includes re-casting the sight rays 530 as each of the subsequent slices becomes the current slice.
  • the first preferred form of the method includes the additional step of performing at least one of zero order inte ⁇ olation, first order inte ⁇ olation, second order inte ⁇ olation, third order inte ⁇ olation, higher order inte ⁇ olation and adaptive combinations thereof.
  • Bilinear inte ⁇ olation as discussed above, is preferred.
  • the inte ⁇ olation is performed among pixels of the compositing buffer (slice 522) adjacent the intersection points 532, in order to provide inte ⁇ olated pixel values for combination with the sample points of the current slice 526.
  • the four voxels (pixels in the compositing buffer) which define the vertices of the square which encloses the intersection point 532 are utilized.
  • inte ⁇ olation For order 2 inte ⁇ olation, the nine closest points, centered about the nearest neighbor are utilized. For order 3 inte ⁇ olation, inner and outer squares are utilized. It will be apparent to those skilled in the art how to adapt other types of inte ⁇ olation to the present method.
  • the first processing slice 522 is normally in front of the buffer and the last processing slice 524 is normally in back of the buffer, such that front-to-back sweeping is performed.
  • front-to-back compositing is also preferably performed, as previously discussed and shown in Figure 19.
  • the preferred method can also be performed with the viewpoint 412 in front, as shown in Figure 43, but utilizing slice 524 as the first processing slice and slice 522 as the last processing slice, such that back-to-front sweeping is instead performed (depicted in
  • the combining is performed from back-to-front with respect to the view point 412, utilizing the back-to-front compositing techniques depicted in Figure 19. It is to be understood that other types of combining techniques rather than compositing can be employed, however, compositing is preferred. Yet further, in the first preferred method, when the view point 412 is in back as depicted in Figure 18, front-to-back sweeping and back-to-front compositing can be performed. In this case, the first processing slice would be slice 522 and the last processing slice would be slice 524. Back-to-front compositing as depicted in Figure 19 would be employed.
  • the back and front with respect to sweeping are defined in dataset coordinates, while back and front with respect to compositing is relative to the eye (view point 412).
  • the view point 412 may be behind the data set, with the first processing slice in the back of the 3-D buffer (slice 524) and the last processing slice in the front of the three dimensional buffer (slice 522).
  • the combining would be performed from front-to-back with respect to the view point 412, using the front-to-back compositing techniques of Figure 19, for example.
  • the sweeping and compositing order options presented in this and the preceding paragraph are referred to hereinafter as "sweeping/compositing order options.”
  • the combining step includes the sub-step of casting a plurality of sight rays 530 from the view point 412 through locations in the adjacent slice 536 associated with the compositing buffer.
  • the locations correspond to grid point locations of pixels within the compositing buffer (voxels 494 within the associated slice).
  • the sight rays 530 intersect the current slice 534 at a plurality of intersection points 538 designated by crosses.
  • the sampling step includes sampling the voxel values from the three dimensional buffer 450 containing the three dimensional volumetric image 404 onto the plurality of intersection points 538 by performing at least one of zero order inte ⁇ olation, first order inte ⁇ olation, second order inte ⁇ olation, third order inte ⁇ olation, higher order inte ⁇ olation and adaptive combinations thereof.
  • the inte ⁇ olation is performed among voxel values associated with voxel locations adjacent the plurality of intersection points 538. Bilinear inte ⁇ olation, as discussed above, is preferred.
  • the step of repeating the sampling and combining steps includes re-casting the sight rays 530 as each of the subsequent slices becomes the current slice.
  • the sampling step includes sampling voxel values of voxel elements 494 from the three dimensional buffer 450 directly onto grid point locations of the current slice 526 of sample points (referring to Figure 43), with the sample points coinciding with the grid points.
  • the first execution of the combining step includes the sub-step of casting a plurality of sight rays 530 from the view point 412 through the grid points of the current slice 526 of sample points, with the interslice ray segments 528 extending from the sample points of the current slice 526 to intersections 532 of the sight rays 530 with the compositing buffer.
  • the third preferred form of the method of the present invention also includes performing one of zero order, first order, second order, third order or higher order inte ⁇ olation and adaptive combinations thereof.
  • the inte ⁇ olation is performed among pixels (corresponding to the voxel elements) of the compositing buffer (slice 522) in order to provide inte ⁇ olated pixel values for combination with the sample points of the current slice 526.
  • the results of the inte ⁇ olation and the combination are stored, with an accumulated offset, in grid point locations of those of the pixels in the compositing buffer which are closest to the intersections 532 of the sight rays 530 with the composit- ing buffer (slice 522).
  • the third preferred method also includes performing either merging of two or more of the interslice ray segments 528 or splitting of at least one of the interslice ray segments 528.
  • Merging of two or more of the interslice ray segments 528 is performed when an attempt is made to store two or more of the results of the inte ⁇ olation and the combination at an identical grid point location. Merging has been discussed extensively above and has been depicted in Figure 41.
  • Splitting is performed when adjacent interslice ray segments 528 diverge beyond a given divergence threshold.
  • the divergence threshold is selected as follows. The divergence can initially be started at zero and when it exceeds a divergence threshold of one, splitting can be performed. This corresponds to an empty cell between any two rays. Alternatively, rays can be started with a distance of one between them, and splitting can be performed when a divergence threshold of two is exceeded. Splitting has been discussed above and will be discussed in further detail below.
  • the third preferred method also includes the step of maintaining the sight rays 530 formed during the first execution of the combining step throughout the sequential sweeping through the subsequent slices.
  • the sight rays are not re-cast when every new slice is traversed. New sight rays are cast, however, when either merging or splitting is performed.
  • a new sight ray is generated from the view point 412 to a preselected location in the neighborhood of the identical grid point location when merging is performed, as depicted in Figure 41.
  • the preselected location will, as noted, be in a neighborhood of the identical grid point where the attempt is made to store two or more of the results of the inte ⁇ olation and the combination.
  • the new sight ray could be cast to the identical grid point location, but can start with an initial offset of plus or minus one-half voxel to the right or left, depending on the accumulated offsets of the two merging rays.
  • two new sight rays are generated from the view point to corresponding preselected locations in the current slice, as discussed below.
  • the preselected locations for generating the two new sight rays may be selected as follows. If the single ray to be split hits exactly between two grid points, the new rays can be started at the adjacent grid point locations. If the single ray hits closer to the left most grid location, one starting point can be shifted to the left with the old ray maintained in the middle with an initial offset. Similarly, if the single ray hits closer to the right, the start point could be shifted to the right, with the old ray in the middle with an initial offset.
  • the third preferred method includes storing the results of the inte ⁇ olation and the combination in an adjacent grid point location when the accumulated offset exceeds a predetermined value, for example, more than halfway to the adjacent grid point, as depicted in Figure 40.
  • the storing is done together with a new accumulated offset value.
  • results are initially stored, at slice 2 in the left most voxel indicated by the dark circle, with an accumulated offset of 0.3.
  • the accumulated offset is 0.6 such that the adjacent grid point, indicated by the black circle, is the appropriate place to store the results. Since a displacement of 1 to the right has occurred, 1 is subtracted from 0.6, yielding a -0.4 offset, which is the new accumulated offset value.
  • merging is preferably performed when the combining step is performed from back to front with respect to view point 412. This case is illustrated in Figure 41.
  • bilinear inte ⁇ olation is used in the inte ⁇ olation step.
  • splitting of interslice ray segments 528 is preferably performed when the combining step is performed from front to back with respect to the view point 412.
  • Figure 47 depicts a form of the third preferred method using splitting with front-to-back compositing and front-to-back processing (i.e. sweeping).
  • sight rays 530 are cast from the view point 412 through grid points (valid voxels indicated by open circles) in the current slice of sample points when a split occurs. Ray splits are indicated by the dotted horizontal lines with arrow heads at either end (items 550). Once again, the "x" marks denote voxels beyond 45°
  • an additional step is included which comprises dividing the three dimensional buffer 450 into a plurality of zones 540 which are separated by zone boundaries 542.
  • merging and splitting are confined to zones 540, and preferably do not occur across zone boundaries 542.
  • the zones 540 and zone boundaries 542 are selected for enhancing image sha ⁇ ness and accuracy.
  • One preferred method of determining the zone boundaries 542 is to form generalized three dimensional pyramids which extend from the view point 412 through the boundaries of pixels (two dimensional consideration of voxels 494) associated with the base plane 408 into the three dimensional buffer 450.
  • the zones 540 in Figure 45 are generalized three dimensional pyramids although Figure 45 is a two dimensional view for simplicity. It is also preferred to perform a step of labeling those voxels 494 which belong to a given one of the zones 540.
  • the encircled voxel elements 494 in Figure 45 could all be given the label " ⁇ " to signify that they belong to the given one of the zones 542 which is designated as zone gamma.
  • the first execution of the combining step includes the sub-step of casting a plurality of sight rays 530 from the view point 412 through locations in the adjacent slice 536 associated with the compositing buffer (refer to Figure 44).
  • the locations correspond to grid point locations of the pixels (two dimensional manifestations of voxel elements 494) within the compositing buffer.
  • the sight rays 530 intersect the current slice 534 at a plurality of intersection points 538.
  • the interslice ray segments 528 extend from the grid point locations of the pixels within the compositing buffer to the intersection points 538.
  • a sampling step includes sampling voxel values from the three dimensional buffer 450 onto the plurality of intersection points 538 by performing one of zero order inte ⁇ olation, first order inte ⁇ olation, second order inte ⁇ olation, third order inte ⁇ olation, higher order inte ⁇ olation and adaptive combinations thereof.
  • the inte ⁇ olation is performed among voxel values associated with voxel locations 494 adjacent the plurality of intersection points 538. Bilinear inte ⁇ olation is preferred.
  • the method further includes the additional step (refer to Figure 40) of storing the results of the inte ⁇ olation and combination, together with an accumulated offset, in grid point locations of the pixels within the compositing buffer.
  • the fourth preferred form of the method further includes performing one of merging of two or more of the interslice ray segments, as shown in Figure 41 and discussed above, or splitting at least one of the interslice ray segments 528.
  • merging is performed when an attempt is made to store two or more of the results of the inte ⁇ olation and combination in an identical grid point location, while splitting is performed when adjacent interslice ray segments 528 diverge beyond a given divergence threshold.
  • the fourth preferred form of the method also includes maintaining the sight rays 530 formed in the first execution of the combining step throughout sequential sweeping through subsequent slices, that is, new sight rays are not normally cast when each subsequent slice is processed. However, new sight rays are cast whenever one of merging and splitting is performed.
  • a new sight ray is generated from the view point 412 to a preselected location in the neighborhood of the identical grid point location (as discussed above); while when splitting is performed, two new sight rays are generated from the view point 412 to corresponding preselected locations in the compositing buffer (as discussed above with respect to the current slice for the third preferred method; the same considerations apply here). Merging and splitting have been discussed in detail above.
  • the fourth preferred form of the method preferably includes storing the results of the inte ⁇ olation and combination in an adjacent grid point location when the accumulated offset exceeds a predetermined value; the storing is done together with a new value for the accumulated offset.
  • the merging is preferably employed when the combining step is performed from back to front with respect to the view point 412, while splitting is preferably employed when the combining step is performed from front to back with respect to the view point 412.
  • Figure 48 depicts a form of the fourth preferred method using splitting with front-to- back compositing and front-to-back processing (i.e., sweeping).
  • two new sight rays 530 are cast from the view point 412 through preselected locations 552 in the compositing buffer slice whenever a splitting operation is carried out.
  • Ray splits are again indicated by the dotted horizontal lines with arrow heads at either end (items 550).
  • an especially preferred form of the fourth method includes the additional step of dividing the three dimensional buffer 450 into a plurality of zones 540 separated by zone boundaries 542.
  • the merging and splitting are confined to individual zones 540 and also to adjacent zones on either side of a given zone 542.
  • merging and splitting, for any given zone 542 do not occur across more than one of the zone boundaries 542 on each side of the given zone 540.
  • the zones and zone boundaries are selected for enhancing image sha ⁇ ness and accuracy and reduction of alaising.
  • the zones and zone boundaries 540, 542 are preferably determined by forming generalized three dimensional pyramids which extend from the view point 412 through boundaries of pixels (two dimensional manifestation of voxels 494) associated with the base plane 408 into the three dimensional buffer 450. Again, those voxels 494 that belong to a given one of the zones 540 can be labeled.
  • perspective projection can be obtained with the view point 412 at a noninfinite distance
  • paralleled projection can be obtained by locating the view point 412 at an effectively infinite distance from the dataset.
  • An apparatus for parallel and perspective real time volume visualization via ray-slice-sweeping is responsive to viewing and processing parameters which define a view point.
  • the apparatus generates a three dimensional volume projection image from the view point.
  • the image has a plurality of pixels.
  • the apparatus includes three dimensional buffer 450; a pixel bus 442 which provides global horizontal communication; a plurality of rendering pipelines 440; and a control unit 454 which can be configured as part of the memory 450.
  • the three dimensional buffer 450 stores a plurality of discrete voxels, each of which has a location and at least one voxel value associated with it.
  • the three dimensional buffer 450 includes a plurality of memory modules 438.
  • the viewing and processing parameters define at least one base plane of the three dimensional buffer 450, as well as first and last processing slices of the three dimensional buffer 450, as discussed above.
  • Each of the rendering pipelines 440 is vertically coupled to a corresponding one of the plurality of memory units 438 and to the pixel bus 442.
  • Each rendering pipeline 440 has horizontal communication with at most its two nearest neighbors, i.e., the neighbor on each side.
  • Each of the rendering pipelines 440 in turn includes at least a first slice unit 544, which can be in the form of a two-dimensional (2-D) first-in-first-out (FIFO) slice buffer.
  • Slice unit 544 has an input coupled to the corresponding one of the plurality of memory units 438 and has an output.
  • the slice unit 544 in addition to being in the form of the FIFO slice buffer, could be an internal FIFO within the memory module 338, or could include special circuitry within the memory module 338 which permits output of two beams in parallel.
  • the at least first slice unit 544 includes a current slice of sample points parallel to the first processing slice (discussed above).
  • the at least first slice unit 544 also receives voxel values from the three dimensional buffer 450 onto the sample points of the current slice to provide sample point values.
  • Each rendering pipeline 440 further includes a compositing unit 464 having an input which is coupled to the output of the at least first slice unit and having an output which is coupled to the pixel bus 442.
  • each rendering pipeline 440 has a two dimensional slice compositing buffer 474 which has a plurality of pixels. Each of the pixels of the two dimensional slice compositing buffer 474 has at least color, transparency (alternatively, opacity) and position associated therewith.
  • the compositing buffer 474 has an input coupled to the output of the compositing unit 464 and has an output coupled back to the input of the compositing unit 464.
  • Each of the rendering pipelines 440 also includes a first bilinear (or other order) inte ⁇ olation unit. The first inte ⁇ olation unit has an input and an output.
  • the input of the first inte ⁇ olation unit can be coupled to the output of the compositing buffer, as for inte ⁇ olation unit 472 in Figure 46.
  • the input of the inte ⁇ olation unit can be coupled to the corresponding one of the plurality of memory units 438, as for inte ⁇ olation unit 546 in Figure 46.
  • the first inte ⁇ olation unit also has an output. The output is preferably coupled to the input of the compositing unit 464 for the case when the input of the first inte ⁇ olation unit is coupled to the output of the compositing buffer, as for inte ⁇ olation unit 472 in Figure 46.
  • the output of the first inte ⁇ olation unit can be coupled to the input of the at least first slice unit 544 when the input of the first inte ⁇ olation unit is coupled to the corresponding one of the plurality of memory units, as for inte ⁇ olation unit 546.
  • the control unit 354 initially designates the first processing slice as the current slice and controls sweeping through subsequent slices of the three dimensional buffer 450, with each slice in turn becoming the current slice, until the last processing slice is reached.
  • the sample point values are combined with the pixels of the compositing buffer 474 in the compositing unit 464, along a plurality of interslice ray segments which extend only between the current slice in the first slice unit 544 and a slice contained in the two dimensional slice compositing buffer 474.
  • the first slice unit 544 receives the voxel values from the three dimensional buffer 450 directly onto grid points of the current slice of sample points, with the sample points coinciding with the grid points.
  • the interslice ray segments extend along a plurality of sight rays cast from the view point through the grid points of the current slice of sample points.
  • the interslice ray segments extend from the sample points of the current slice in the first slice unit 544 to the slice contained in the two dimensional slice compositing buffer 474.
  • the interslice ray segments intersect the slice contained in the two dimensional slice compositing buffer 474 at a plurality of intersection points.
  • the inte ⁇ olation unit has its input coupled to the output of the compositing buffer 474 and its output coupled to the input of the compositing unit 464. That is, the configuration of inte ⁇ olation unit designated as 472 in Figure 46 is employed.
  • the inte ⁇ olation unit 472 receives signals associated with pixels of the two dimensional slice compositing buffer 474 which are adjacent the intersection points, and provides inte ⁇ olated pixel values for combination with the sample points of a current slice in the compositing unit 464.
  • the control unit 454 recasts the sight rays as each of the subsequent slices becomes the current slice.
  • each of the rendering pipelines 440 further includes a second slice unit 548 having an input coupled to the output of the first slice unit 544 and having an output.
  • the second slice unit can be a FIFO or special configuration of memory 438, as set forth above with respect to the first slice unit.
  • Each pipeline 440 also preferably contains a gradient computation unit 468 having an input coupled to the output of the second slice unit 548 and having an output; and each rendering pipeline 440 preferably further contains a shading unit 466 having input coupled to the output of the gradient computation unit 468 and having an output coupled to the input of the compositing unit 464.
  • the second slice unit 548 is used for gradient computation in conjunction with the gradient computation unit 468.
  • the first inte ⁇ olation unit has its input coupled to the three dimensional buffer 450 and its output coupled to the input of the first slice unit 544, i.e., the inte ⁇ olation unit is configured as item 546 in Figure 46.
  • the interslice ray segments extend along a plurality of sight rays cast from the view point through locations in the slice contained in the 2-D slice compositing buffer 474. The locations correspond to grid point locations of the pixels within the compositing buffer 474, and the sight rays intersect the current slice at a plurality of intersection points.
  • the first inte ⁇ olation unit 546 receives voxel values associated with voxel locations adjacent the plurality of intersection points and inte ⁇ olates among the voxel values to provide inte ⁇ olated voxel values associated with the plurality of intersection points.
  • the control unit 454 recasts the sight rays as each of the subsequent slices becomes the current slice.
  • each of the plurality of rendering pipelines 440 further includes a second slice unit 548 having an input coupled to the output of the first slice unit 544 and having an output; a gradient computation unit 468 having an input coupled to the output of the second slice unit 548 and having an output; and a shading unit 466 having an input coupled to the output of the gradient computation unit 468 and having an output coupled to the input of the compositing unit 464.
  • the second slice unit 548 can be a FIFO or specially configured memory, as discussed above.
  • each of the plurality of rendering pipelines 440 preferably also includes a second inte ⁇ olation unit (preferably bilinear) having an input coupled to the output of the compositing buffer 474 and having an output coupled to the input of the compositing unit 464, i.e., inte ⁇ olation unit 472 is preferably also used along with inte ⁇ olation unit 546 in the second preferred form of apparatus.
  • the second slice unit 548 is employed for gradient estimation in conjunction with the gradient computation unit 468.
  • the first slice unit 544 receives the voxel values from the three dimensional buffer 450 directly onto grid points of the current slice of sample points, with the sample points coinciding with the grid points.
  • the interslice ray segments extend along a plurality of sight rays cast from the view point through the grid points of the current slice of sample points.
  • the interslice ray segments extend from the sample points of the current slice in the slice unit 544 to the slice contained in the two dimensional slice compositing buffer 474, with the interslice ray segments intersecting the slice contained in the two dimensional slice compositing buffer 474 at a plurality of intersection points.
  • the first inte ⁇ olation unit (preferably bilinear) has its input coupled to the output of the compositing buffer 474 and its output coupled to the input of the compositing unit 464.
  • the first inte ⁇ olation unit is configured as item 472 in Figure 46.
  • the first inte ⁇ olation unit 472 receives signals associated with pixels of the two dimensional slice compositing buffer 474 which are adjacent the intersection points and provides inte ⁇ olated pixel values for combination with the sample points of the current slice within the compositing unit 464.
  • Results of the combination of the inte ⁇ olated pixel values and the sample points of the current slice are stored, with an accumulated offset, in grid point locations of those of the pixels in the compositing buffer 474 which are closest to the intersection of the sight rays with the compositing buffer 474.
  • control unit 454 enables one of merging of two or more of the interslice ray segments ray segments or splitting of at least one of the interslice ray segments. It is to be understood that the control unit 454 provides appropriate controls to, for example, the compositing unit 464 so that the compositing unit 464 performs the merging or splitting. Throughout the application, this meaning is included when it is stated that the control unit enables a particular function. As discussed above with respect to the method, merging is performed when an attempt is made to store two or more of the results of the inte ⁇ olation and the combination in an identical grid point location. Splitting is performed when adjacent interslice ray segments diverge beyond a divergence threshold, determined as set forth above with respect to the method.
  • control unit 454 maintains the sight rays throughout the sequential sweeping through the subsequent slices, except when one of the merging and the splitting is performed.
  • the control unit 454 generates a new sight ray from the view point to a preselected location in the neighborhood of the identical grid point location when merging is performed and generates two new sight rays from the view point to corresponding preselected locations in the current slice when the splitting is performed.
  • the preselected locations are determined as set forth above with respect to the method.
  • the control unit 454 stores the results of the inte ⁇ olation and combination in an adjacent one of the grid point locations when the accumulated offset exceeds a predetermined value (also as set forth above with respect to the method). The storing is done together with a new accumulated offset value.
  • the control unit 454 divides the three dimensional buffer 350 into a plurality of zones separated by zone boundaries of the type described above, and the merging and splitting are confined to the zones and do not occur across the zone boundaries.
  • the zone and zone boundaries are selected for enhancing image sha ⁇ ness and accuracy, as set forth above.
  • the first inte ⁇ olation unit has its input coupled to the three dimensional buffer 450 and its output coupled to the input of the first slice unit 544, i.e., the first inte ⁇ olation unit (preferably bilinear) is connected in accordance with item 546 in Figure 46.
  • interslice ray segments extend along a plurality of sight rays cast from the view point through locations in the slice contained in the two dimensional slice compositing buffer 474, with the locations corresponding to grid point locations of the pixels within the compositing buffer 474.
  • the sight rays intersect the current slice at a plurality of intersection points.
  • the first inte ⁇ olation unit 546 receives voxel values associated with voxel locations adjacent the plurality of intersection points and inte ⁇ olates among the voxel values to provide inte ⁇ olated voxel values associated with the plurality of intersection points for combination with the pixels of the compositing buffer 474. Results of the combination are stored, with an accumulated offset, in the grid point locations of the pixels within the compositing buffer 474.
  • control unit 454 enables (see discussion above) one of merging and splitting of the interslice ray segments. Specifically, merging of two or more of the interslice ray segments is performed when an attempt is made to store two or more of the results of the inte ⁇ olation and combination at an identical grid point location. Splitting of at least one of the interslice ray segments is performed when adjacent interslice ray segments diverge beyond a divergence threshold (discussed above with respect to the method). Further, the control unit 454 maintains the sight rays throughout the sequential sweeping through the subsequent slices, except when the merging or splitting is performed.
  • control unit 454 generates a new sight ray from the view point to a preselected location in the neighborhood of the identical grid point location when merging is performed and generates two new sight rays from the view point to corresponding preselected locations in the compositing buffer 474 when splitting is performed.
  • control unit 454 stores the results of the inte ⁇ olation and combination in an adjacent one of the grid point locations when the accumulated offset exceeds a predetermined value. The storing is done together with a new accumulated offset value. Further details are provided above in the discussion of the fourth preferred method.
  • the control unit 454 divides the three dimensional buffer 450 into a plurality of zones separated by zone boundaries, as described above.
  • the merging and splitting is confined to individual zones and, in the case of the fourth preferred apparatus, to adjacent zones (the zone immediately on either side) and does not occur, for a given zone, across more than one of the zone boundaries on each side of the given zone.
  • the zones and zone boundaries are selected for enhancing image sha ⁇ ness and accuracy and for reduction of aliasing, as set forth above with respect to the fourth preferred method.
  • the apparatus in accordance with the present invention can include a global feedback connection 548 between pixel bus 442 and 3-D buffer 450, as shown in Figure 20.
  • Global feedback connection 548 forms a global feedback loop.
  • the global feedback loop is configured for feedback from the pixel bus 442 to the three dimensional buffer 450, and subsequently to any of the intermediate stages 544, 546, 548, 468, 466, 464, 474 and 472, for example, in the rendering pipelines 440. Feedback to any of the intermediate stages depicted in Figure 22 is also possible.
  • Global feedback enables reuse of intermediate results and enhances the flexibility of the present apparatus and method.
  • any of the methods set forth above can include a method step of providing global feedback from the pixel bus 442 to the three dimensional buffer 450, and subsequently to any of the intermediate stages.
  • the global feedback can be provided for pu ⁇ oses of changing the order of computation in a given pipeline 440.

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Computer Graphics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Generation (AREA)
  • Image Processing (AREA)

Abstract

L'appareil de l'invention comprend une mémoire-tampon tridimensionnelle (438); un bus à pixels (442); plusieurs pipelines de rendu (440); et une unité de commande (454). Chaque pipeline de rendu comprend une première unité à tranches (544); une unité de composition d'images (464); un tampon de composition bidimensionnelle par tranche (474); et une première unité d'interpolation (472 ou 546). Les valeurs des points échantillons sont combinées aux pixels du tampon de composition d'images, le long de plusieurs segments de rayon intertranche (410) qui s'étendent seulement entre une tranche en cours contenue dans l'unité à tranches et une tranche contenue dans le tampon de composition d'images bidimensionnelle par tranche. Les gradients sont également calculés au niveau des positions des voxels, ce qui améliore la précision.
PCT/US1998/007405 1997-04-15 1998-04-13 Appareil et procede pour la visualisation de volume en temps reel, en parallele et en perspective WO1998047105A1 (fr)

Priority Applications (4)

Application Number Priority Date Filing Date Title
JP10525927A JPH11508386A (ja) 1997-04-15 1998-04-13 平行法及び遠近法によりボリュームを実時間で視覚化する装置及び方法
EP98918164A EP0992023A4 (fr) 1997-04-15 1998-04-13 Appareil et procede pour la visualisation de volume en temps reel, en parallele et en perspective
AU71139/98A AU732652B2 (en) 1997-04-15 1998-04-13 Apparatus and method for parallel and perspective real-time volume visualization
CA002285966A CA2285966A1 (fr) 1997-04-15 1998-04-13 Appareil et procede pour la visualisation de volume en temps reel, en parallele et en perspective

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US4301397P 1997-04-15 1997-04-15
US60/043,013 1997-04-15
US08/910,575 1997-08-01
US08/910,575 US5847711A (en) 1994-09-06 1997-08-01 Apparatus and method for parallel and perspective real-time volume visualization

Publications (1)

Publication Number Publication Date
WO1998047105A1 true WO1998047105A1 (fr) 1998-10-22

Family

ID=26719915

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US1998/007405 WO1998047105A1 (fr) 1997-04-15 1998-04-13 Appareil et procede pour la visualisation de volume en temps reel, en parallele et en perspective

Country Status (5)

Country Link
EP (1) EP0992023A4 (fr)
JP (1) JPH11508386A (fr)
AU (1) AU732652B2 (fr)
CA (1) CA2285966A1 (fr)
WO (1) WO1998047105A1 (fr)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0950986A2 (fr) * 1998-04-13 1999-10-20 Mitsubishi Denki Kabushiki Kaisha Système de rendu parallèle de volumes avec module de rééchantillonnage pour projections parallèles et en perspective
EP1001375A2 (fr) * 1998-11-12 2000-05-17 Mitsubishi Denki Kabushiki Kaisha Recadrage avec pipeline pour manipuler des ensembles de données de volume en temps réel
WO2000033257A1 (fr) * 1998-11-27 2000-06-08 Algotec Systems Ltd. Procede pour former un rendu en perspective a partir d'un espace de voxels
EP1054351A2 (fr) * 1999-05-20 2000-11-22 Mitsubishi Denki Kabushiki Kaisha Méthode et appareil de classification de valeurs d'intensité de données volumétriques à l'aide d'une table de consultation
EP1209629A1 (fr) * 2000-11-28 2002-05-29 TeraRecon, Inc., A Delaware Corporation Fin de rayon précoce dans un système de rendu volumique parallèle en pipeline
WO2002069274A1 (fr) * 2001-02-27 2002-09-06 Siemens Corporate Research, Inc. Algorithme de projection de voxels par decoupage-torsion econome en memoire
EP1315465A2 (fr) * 2000-09-07 2003-06-04 Cbyon, Inc. Systeme et procede radioscopiques virtuels
US7054794B2 (en) 2000-06-01 2006-05-30 Richard David Brunt Color visualisation system
US9002090B2 (en) 2010-11-25 2015-04-07 Koninklijke Philips N.V. Forward projection apparatus
CN113409436A (zh) * 2021-06-02 2021-09-17 东北大学 一种菱形像素排列的体绘制方法

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6512517B1 (en) * 1998-11-12 2003-01-28 Terarecon, Inc. Volume rendering integrated circuit
JP2001022950A (ja) * 1999-05-20 2001-01-26 Mitsubishi Electric Inf Technol Center America Inc ボリュームレンダリンググラフィックボード
JP4493151B2 (ja) * 2000-04-03 2010-06-30 株式会社日立メディコ 画像表示装置
US20130222383A1 (en) * 2010-11-12 2013-08-29 Hitachi Medical Corporation Medical image display device and medical image display method

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4827413A (en) * 1987-06-16 1989-05-02 Kabushiki Kaisha Toshiba Modified back-to-front three dimensional reconstruction algorithm
US5381518A (en) * 1986-04-14 1995-01-10 Pixar Method and apparatus for imaging volume data using voxel values
US5412764A (en) * 1990-06-22 1995-05-02 Kabushiki Kaisha Toshiba Three-dimensional image display apparatus using numerical projection
US5557711A (en) * 1990-10-17 1996-09-17 Hewlett-Packard Company Apparatus and method for volume rendering
US5594844A (en) * 1993-01-26 1997-01-14 Hitachi, Ltd. Three dimensional view using ray tracing through voxels subdivided numerically using object based parameters
US5594842A (en) * 1994-09-06 1997-01-14 The Research Foundation Of State University Of New York Apparatus and method for real-time volume visualization

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5381518A (en) * 1986-04-14 1995-01-10 Pixar Method and apparatus for imaging volume data using voxel values
US4827413A (en) * 1987-06-16 1989-05-02 Kabushiki Kaisha Toshiba Modified back-to-front three dimensional reconstruction algorithm
US5412764A (en) * 1990-06-22 1995-05-02 Kabushiki Kaisha Toshiba Three-dimensional image display apparatus using numerical projection
US5557711A (en) * 1990-10-17 1996-09-17 Hewlett-Packard Company Apparatus and method for volume rendering
US5594844A (en) * 1993-01-26 1997-01-14 Hitachi, Ltd. Three dimensional view using ray tracing through voxels subdivided numerically using object based parameters
US5594842A (en) * 1994-09-06 1997-01-14 The Research Foundation Of State University Of New York Apparatus and method for real-time volume visualization

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP0992023A4 *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0950986A3 (fr) * 1998-04-13 2001-01-10 Mitsubishi Denki Kabushiki Kaisha Système de rendu parallèle de volumes avec module de rééchantillonnage pour projections parallèles et en perspective
EP0950986A2 (fr) * 1998-04-13 1999-10-20 Mitsubishi Denki Kabushiki Kaisha Système de rendu parallèle de volumes avec module de rééchantillonnage pour projections parallèles et en perspective
EP1001375A2 (fr) * 1998-11-12 2000-05-17 Mitsubishi Denki Kabushiki Kaisha Recadrage avec pipeline pour manipuler des ensembles de données de volume en temps réel
EP1001375A3 (fr) * 1998-11-12 2002-01-09 TeraRecon, Inc., A Delaware Corporation Recadrage avec pipeline pour manipuler des ensembles de données de volume en temps réel
WO2000033257A1 (fr) * 1998-11-27 2000-06-08 Algotec Systems Ltd. Procede pour former un rendu en perspective a partir d'un espace de voxels
US6999078B1 (en) 1998-11-27 2006-02-14 Algotec Systems Ltd. Method of forming a perspective rendering from a voxel space
EP1054351A3 (fr) * 1999-05-20 2003-06-18 TeraRecon, Inc., A Delaware Corporation Méthode et appareil de classification de valeurs d'intensité de données volumétriques à l'aide d'une table de consultation
EP1054351A2 (fr) * 1999-05-20 2000-11-22 Mitsubishi Denki Kabushiki Kaisha Méthode et appareil de classification de valeurs d'intensité de données volumétriques à l'aide d'une table de consultation
US7054794B2 (en) 2000-06-01 2006-05-30 Richard David Brunt Color visualisation system
EP1315465A2 (fr) * 2000-09-07 2003-06-04 Cbyon, Inc. Systeme et procede radioscopiques virtuels
EP1209629A1 (fr) * 2000-11-28 2002-05-29 TeraRecon, Inc., A Delaware Corporation Fin de rayon précoce dans un système de rendu volumique parallèle en pipeline
US6570952B2 (en) 2001-02-27 2003-05-27 Siemens Corporate Research, Inc. Memory efficient shear-warp voxel projection algorithm
WO2002069274A1 (fr) * 2001-02-27 2002-09-06 Siemens Corporate Research, Inc. Algorithme de projection de voxels par decoupage-torsion econome en memoire
US9002090B2 (en) 2010-11-25 2015-04-07 Koninklijke Philips N.V. Forward projection apparatus
CN113409436A (zh) * 2021-06-02 2021-09-17 东北大学 一种菱形像素排列的体绘制方法
CN113409436B (zh) * 2021-06-02 2023-10-13 东北大学 一种菱形像素排列的体绘制方法

Also Published As

Publication number Publication date
JPH11508386A (ja) 1999-07-21
EP0992023A1 (fr) 2000-04-12
CA2285966A1 (fr) 1998-10-22
AU7113998A (en) 1998-11-11
EP0992023A4 (fr) 2001-04-18
AU732652B2 (en) 2001-04-26

Similar Documents

Publication Publication Date Title
US5847711A (en) Apparatus and method for parallel and perspective real-time volume visualization
US5594842A (en) Apparatus and method for real-time volume visualization
EP0447227B1 (fr) Méthode pour générer des adresses des primitives graphiques mémorisées dans RIP Maps
US6646639B1 (en) Modified method and apparatus for improved occlusion culling in graphics systems
US6756993B2 (en) Methods and apparatus for rendering images using 3D warping techniques
Heidrich et al. View-independent environment maps
US4967392A (en) Drawing processor for computer graphic system using a plurality of parallel processors which each handle a group of display screen scanlines
US6674430B1 (en) Apparatus and method for real-time volume processing and universal 3D rendering
AU732652B2 (en) Apparatus and method for parallel and perspective real-time volume visualization
US5684935A (en) Rendering and warping image generation system and method
US6362818B1 (en) System and method for reducing the rendering load for high depth complexity scenes on a computer graphics display
US6219061B1 (en) Method for rendering mini blocks of a volume data set
EP1054348A2 (fr) Circuit intégré de rendu de volume
AU757621B2 (en) Apparatus and method for real-time volume processing and universal 3D rendering
US5757374A (en) Method and apparatus for performing texture mapping
US20020005850A1 (en) Super-sampling and gradient estimation in a ray-casting volume rendering system
JP2000348202A (ja) ボクセルを有するボリューム・データ・セットのずらし−ワープ・レンダリング方法、ボリューム・データ・セットのレンダリング装置
US7106352B2 (en) Automatic gain control, brightness compression, and super-intensity samples
US5719598A (en) Graphics processor for parallel processing a plurality of fields of view for multiple video displays
Pfister et al. Cube-3: A real-time architecture for high-resolution volume visualization
Bitter et al. A ray-slice-sweep volume rendering engine
US5740344A (en) Texture filter apparatus for computer graphics system
US5732248A (en) Multistep vector generation for multiple frame buffer controllers
Popescu Forward rasterization: A reconstruction algorithm for image-based rendering
EP1054384A2 (fr) Méthode et dispositif de traduction et de coopération avec des adresses de mémoire de voxels

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref country code: JP

Ref document number: 1998 525927

Kind code of ref document: A

Format of ref document f/p: F

AK Designated states

Kind code of ref document: A1

Designated state(s): AL AM AT AU AZ BA BB BG BR BY CA CH CN CU CZ DE DK EE ES FI GB GE GH GM GW HU ID IL IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MD MG MK MN MW MX NO NZ PL PT RO RU SD SE SG SI SK SL TJ TM TR TT UA UG UZ VN YU ZW

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): GH GM KE LS MW SD SZ UG ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE BF BJ CF CG CI CM GA GN ML MR NE SN TD TG

DFPE Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101)
121 Ep: the epo has been informed by wipo that ep was designated in this application
ENP Entry into the national phase

Ref document number: 2285966

Country of ref document: CA

Ref country code: CA

Ref document number: 2285966

Kind code of ref document: A

Format of ref document f/p: F

WWE Wipo information: entry into national phase

Ref document number: 71139/98

Country of ref document: AU

WWE Wipo information: entry into national phase

Ref document number: 1998918164

Country of ref document: EP

REG Reference to national code

Ref country code: DE

Ref legal event code: 8642

WWP Wipo information: published in national office

Ref document number: 1998918164

Country of ref document: EP

WWG Wipo information: grant in national office

Ref document number: 71139/98

Country of ref document: AU

WWW Wipo information: withdrawn in national office

Ref document number: 1998918164

Country of ref document: EP