CN105005998A - Cerebrovascular image segmentation method based on multi-angle serialized image space feature point set - Google Patents

Abstract

The invention discloses a cerebrovascular image segmentation method based on a multi-angle serialized image space feature point set, comprising the following steps: S1, registering a live image and a mask image of cerebral vessels; S2, extracting a geometric feature point set of a serialized subtraction image; S3, locally adjusting the positions of the feature points at the edges of the blood vessels in the serialized subtraction image; S4, adopting a spatial rotating coordinate system to remove the feature points in non-vascular positions in the subtraction image; S5, determining the adaptive segmentation threshold of the serialized subtraction image; and S6, based on region growing and vascular image segmentation of adaptive threshold and by taking the feature points obtained in S4 as seed points and the adaptive segmentation threshold obtained in S5 as the standard of the law of growth, adopting a region growing algorithm to segment the serialized subtraction image to obtain a pure cerebrovascular image.

Description

A kind of cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set

Technical field

The present invention relates to technical field of image processing, particularly relate to a kind of cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set.

Background technology

Cerebrovascular image is obtained by DSA technology, doctor can be helped to diagnose patient, but the breathing due to patient such as to swallow at the action, obvious artifact and noise can be there is, have impact on observation and the diagnosis of doctor, therefore, it is current based on urgent problem in the blood vessel diagnosis and treatment of DSA technology for extracting vascular tissue's image accurately.Prior art comprises manual segmentation method, method for registering, based on the method for model.The method of manual segmentation needs manually to carry out alternately, the segmentation result difference that different manual operations obtains is very large, also need the time of at substantial, larger noise is comprised in the blood-vessel image obtained by method for registering, affect the diagnosis of doctor, adopt the method based on model, need a template image, only have accurate template just can obtain good vessel segmentation.

Summary of the invention

According to prior art Problems existing, the invention discloses a kind of cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set, comprise the following steps:

S1: registration is carried out to cerebrovascular flapper image and frisket image: determine geometric properties point on flapper image and frisket image, find out the feature point pairs of coupling, by the relation between the feature point pairs of coupling, process acquisition serializing subtraction image is carried out to frisket image;

S2: the geometric properties point set of abstraction sequence subtraction image;

S3: local location adjustment is carried out to the unique point at serializing subtraction image medium vessels edge: the position being adjusted unique point by gray value gradient, the unique point of vessel boundary is moved to Ink vessel transfusing;

S4: adopt Space Rotating coordinate system model to reject the unique point of non-vascular position in subtraction image: to set up a Space Rotating coordinate system model, utilize the positional information of unique point on adjacent subtraction image, obtain the theoretical position of unique point in other subtraction image, according to the position of unique point in other subtraction images, the mode of voting is taked to reject the unique point of non-vascular position;

S5: the adaptivenon-uniform sampling threshold value determining serializing subtraction image;

S6: the blood-vessel image segmentation based on region growing and adaptive threshold: using the unique point that obtains in S4 as Seed Points, using the adaptivenon-uniform sampling threshold value that obtains in the S5 standard as Growing law, algorithm of region growing sequence of partitions subtraction image is adopted to obtain pure cerebrovascular image.

In S1 specifically in the following way:

S11: utilize SURF algorithm to determine geometric properties point respectively on corresponding flapper image and frisket image, calculate the proper vector of unique point;

S12: utilize FLANN adaptation to carry out proper vector coupling, obtain some matching double points;

S13: to the point matched to carrying out screening eliminating error coupling, calculate the minor increment between matching double points, the point that a reservation distance is less than the minor increment three times between matching double points is right;

S14: adopt RANSAC algorithm to remove error matching points pair, obtain the transformation matrix between two width images, obtain the frisket image after registration by transformation matrix, the frisket image after flapper image and registration is directly subtracted shadow, obtains serializing subtraction image.

In S2 specifically in the following way:

S21: utilize SIFT algorithm to extract geometric properties point set on serializing subtraction image He on frisket image;

S22: carry out upper and lower Fragmentation when obtaining the unique point of subtraction image to image, get 1/3rd of image as upper picture, all the other are as lower picture, the operation upper picture being carried out strengthening to contrast obtains more unique point;

S23: calculate the Euclidean distance between unique point in frisket image and the unique point in subtraction image, if the Euclidean distance between the unique point in the unique point in subtraction image and frisket image is less than 50, then by this unique point, it is rejected.

In S3 specifically in the following way:

S31: judge in subtraction image, whether feature point set is all processed complete, and if it is computation process terminates, if not the unique point that then taking-up one is not processed from feature point set;

S32: judge whether above-mentioned not processed unique point is the point that in 8 fields, gray-scale value is minimum, if it is continues to judge next not processed unique point;

S33: if be judged as in S32 otherwise calculate the gradient on this unique point 8 directions, using point maximum for gradient as new unique point;

S34: repeat above-mentioned steps, until feature point set is all disposed in subtraction image.

In S4 specifically in the following way:

S41: choose a unique point P from current subtraction image ₁(x ₁, y ₁), calculate and P in the adjacent subtraction image in current subtraction image ₁meet the unique point P of following relation ₂(x ₂, y ₂):

|x ₂-x ₁|≤4

|y ₂-y ₁|≤2

S42: according to P ₁horizontal ordinate x ₁, P ₂horizontal ordinate x ₂and P ₁and P ₂between angle [alpha]=2 ° calculate unique point P ₁distance r to rotation central axis:

$r=\frac{\sqrt{{x_2}^2-2x_1{x}_2\cos \mathrm{\α}+{x_1}^2}}{\sin \mathrm{\α}}$

S43: calculate unique point P ₁the theoretical position P projected in other subtraction images _n: set when the sequence number of the subtraction image A of pre-treatment is as i _1,, the sequence number of other subtraction images B is i _n, the angle [alpha] of the rotation between two width subtraction images is obtained by the difference of sequence number:

α＝|2×(i _n-i ₁)|；

S44: obtain unique point P according to the above-mentioned r that calculates and anglec of rotation α ₁the position p of the projection in current subtraction image B _n(x _n, y _n);

x _n＝x ₁±r×cosα

y _n＝y ₁

S45: search the unique point (x whether existing and meet following requirement from subtraction image B _m, y _m), if existed, give unique point P ₁throw a ticket, otherwise do not vote;

|x _m-x _n|≤4

|y _m-y _n|≤2

S46: setting threshold value, rejects number of votes obtained and obtains feature point set lower than after the unique point of this threshold value.

In S5 specifically in the following way:

S51: carry out gaussian filtering operation to subtraction image, removes noise, adopts the mode as described in S22 to carry out burst process to image, the adaptivenon-uniform sampling threshold value of picture in calculating:

Centered by pixel P (x, y), get a w _up× w _upwindow, wherein w _upfor odd number;

Gray scale expectation value E in calculation window _local(x, y);

$E_{local}\left(x,y\right)=\left(\underset{i=x-w_{up}/2}{\overset{x+w_{up}/2}{\Σ}}\underset{j=y-w_{up}/2}{\overset{y+w_{up}/2}{\Σ}}P\left(i,j\right)\right)/{w_{up}}^2$

Intensity profile standard deviation sigma (x, y) in calculation window

$\mathrm{\σ}\left(x,y\right)=\sqrt{\underset{i=x-w_{up}/2}{\overset{x+w_{up}/2}{\Σ}}\underset{j=y-w_{up}/2}{\overset{y+w_{up}/2}{\Σ}}{\left(P\left(i,j\right)-E_{local}\left(x,y\right)\right)}^2}/w_{up}$

According to intensity profile standard deviation sigma (x, y) in the gray scale expectation value Elocal (x, y) in window and window, definite threshold T _up(x, y):

T _up(x,y)＝E _local(x,y)+k _up×σ(x,y)

Wherein: k _upspan is 0.1-0.3, w _upspan is 6-11;

Calculate the adaptivenon-uniform sampling threshold value of lower picture:

Calculate the integrogram of lower picture, in integrogram, the value of any point I (x, y) equals the gray-scale value sum from the image upper left corner to all points in the rectangular area that this point is formed, I represents integral image, G represents subtraction image, wherein 0≤i≤x, 0≤j≤y.

I(x,y)＝sum(G(i,j))

Centered by pixel P (x, y), get a w _down× w _downwindow, wherein w _downfor odd number,

The mean value m (x, y) of the pixel in window is obtained, wherein x-w by integrogram _down/ 2≤i≤x+w _down/ 2, y-w _down/ 2≤j≤y+w _down/ 2.

$\begin{array}{c}Sum\left(G\left(i,j\right)\right)=I\left(x-\frac{w_{down}}{2},y-\frac{w_{down}}{2}\right)+I\left(x+\frac{w_{down}}{2},y+\frac{w_{down}}{2}\right)-\\ \left(I\left(x+\frac{w_{down}}{2},y-\frac{w_{down}}{2}\right)+I\left(x-\frac{w_{down}}{2},y+\frac{w_{down}}{2}\right)\right)\end{array}$

m(x,y)＝Sum(G(i,j))/w _down ²

According to the average gray value m (x, y) in window, definite threshold T _down(x, y):

T _down(x,y)＝k _down×m(x,y)

Wherein: k _downspan is 0.7-0.9, w _downspan is 45-61.

In S6 specifically in the following way:

S61: adopt in S4 and obtain the initial seed point of unique point as algorithm of region growing, create an assistant images identical with subtraction image size, color is set to white, this Seed Points not processed mistake when the color of this seed point location is white on assistant images;

S62: the Seed Points concentrating taking-up one untreated from Seed Points, the color of this seed point location is set to black by assistant images, travels through the pixel in 8 neighborhoods of this Seed Points;

S63: the color obtaining the pixel correspondence position in 8 neighborhoods of current seed point from assistant images, if the color of the pixel in neighborhood on assistant images is white, then continue the pixel in the next neighborhood of process, if the color of the pixel in neighborhood on assistant images is black, then compare the size of the gray-scale value of this pixel and the adaptivenon-uniform sampling threshold value calculated in previous step of current seed point, if the gray-scale value of this pixel is less than adaptivenon-uniform sampling threshold value, then this pixel is joined Seed Points to concentrate, continue next pixel in process neighborhood,

S64: repeat above-mentioned steps, add to come in until no longer include new Seed Points;

S65: a newly-built image identical with subtraction image size, is mapped in newly-built image the pure cerebrovascular image after obtaining segmentation by the gray-scale value of the position of Seed Points corresponding in seed point set and subtraction image.

Owing to have employed technique scheme, cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set provided by the invention, effective these artifacts of removal can be had, blood vessel is split from image, obtain continuous print blood-vessel image clearly, the algorithm that the present invention proposes is simple and efficient, travelling speed is fast, full-automatic, effectively blood vessel can be split from the background of complexity, also there is good segmentation result to tiny capillary, blood-vessel image accurately can be provided for the clinical operation of DSA interventional therapy.

Accompanying drawing explanation

In order to be illustrated more clearly in the embodiment of the present application or technical scheme of the prior art, be briefly described to the accompanying drawing used required in embodiment or description of the prior art below, apparently, the accompanying drawing that the following describes is only some embodiments recorded in the application, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to these accompanying drawings.

Fig. 1 is the process flow diagram of method disclosed in the present invention;

Fig. 2 is the schematic diagram of image sequence acquisition process in the present invention;

Fig. 3 is the schematic diagram of flapper image in the present invention;

Fig. 4 is the schematic diagram of frisket image in the present invention;

Fig. 5 is the schematic diagram of flapper image and frisket Image Feature Point Matching in the present invention;

Fig. 6 is the comparison diagram of subtraction image before and after coupling in the present invention;

Fig. 7 is that in the present invention, unique point carries out the comparison diagram before and after local location adjustment;

Fig. 8 is the schematic diagram of rotating coordinate system model in space in the present invention;

Fig. 9 is the front and back contrast schematic diagram of the unique point rejecting non-vascular position in the present invention;

Figure 10 is the schematic diagram determining adaptive threshold in the present invention;

Figure 11 is the schematic diagram of integrogram in the present invention;

Figure 12 is the schematic diagram of algorithm of region growing in the present invention;

Figure 13 is the blood-vessel image after the final segmentation obtained in the present invention.

Embodiment

For making technical scheme of the present invention and advantage clearly, below in conjunction with the accompanying drawing in the embodiment of the present invention, clear complete description is carried out to the technical scheme in the embodiment of the present invention:

A kind of cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set as shown in Figure 1, in implementation process, image sequence acquisition process schematic is as Fig. 2, the upper left corner is image capture device, the upper right corner is the schematic diagram gathering DSA image sequence from different perspectives, equipment is by rotating C arm, frisket and flapper image sequence is obtained after gathering, then " by " process extract minutiae collection is carried out, then introduce rotating coordinate system and carry out rejecting dirty data, finally carry out dividing processing.Method concrete steps disclosed by the invention are as follows:

S1: registration is carried out to cerebrovascular flapper image and frisket image, remove most motion artifacts and noise, be beneficial to follow-up cutting operation: on flapper image and frisket image, determine geometric properties point, find out the feature point pairs of coupling, by the relation between the feature point pairs of coupling, process acquisition serializing subtraction image is carried out to frisket image.Specifically in the following way:

S11: utilize SURF algorithm to determine geometric properties point respectively on corresponding flapper image and frisket image, calculate the proper vector of unique point; As shown in Figure 3, frisket image as shown in Figure 4 for flapper image.

S12: utilize FLANN adaptation to carry out proper vector coupling, obtain some matching double points; As shown in Figure 5, two points at the two ends of line segment are the feature point pairs of coupling to matching result.

S13: to the point matched to carrying out screening eliminating error coupling, calculate the minor increment between matching double points, the point that a reservation distance is less than the minor increment three times between matching double points is right;

S14: adopt RANSAC algorithm to remove error matching points pair, obtain the transformation matrix between two width images, obtain the frisket image after registration by transformation matrix, the frisket after flapper image and registration is directly subtracted shadow, obtains serializing subtraction image.As shown in Figure 6, Fig. 6 (a) is the image directly subtracting movie queen to subtraction image, and Fig. 6 (b) is the image carrying out subtracting shadow after registration, can see, subtraction image after registration is more clear, and noise is fewer, particularly the region at tooth place.

S2: the geometric properties point set of abstraction sequence subtraction image:

S21: utilize SIFT algorithm to extract geometric properties point set on subtraction image He on frisket image;

S22: upper and lower Fragmentation is carried out to image when obtaining subtraction image unique point, in DSA cerebrovascular image, top and the bottom difference is obvious, upper part is mostly capillary, the gray-value variation of pixel is not obvious, and the blood vessel of lower part is all thicker, when extraction SIFT feature point in order to the capillary in upper slice is extracted as much as possible, need to carry out pretreatment operation to upper slice, get 1/3rd of image as upper picture, all the other are as lower picture, and the operation upper picture being carried out strengthening to contrast obtains more unique point;

S23: the unique point in calculating frisket image and the Euclidean distance between the unique point in subtraction image: in subtraction image, the edge gradient change of profile is obvious, having some unique points is positioned on the edge of profile, and these unique points are unique points of the mistake not on blood vessel, the profile of frisket image and the profile of subtraction image are intimate consistent, and frisket image is there is not blood vessel, the SIFT feature point set of frisket image is extracted at identical threshold condition, by doing difference to the feature point set of subtraction image and frisket characteristics of image point set, the unique point being positioned at contour edge in subtraction image can be rejected.If the Euclidean distance between the unique point in the unique point in subtraction image and frisket image is less than 50, then by this unique point, it is rejected.

S3: local location adjustment is carried out to the unique point at serializing subtraction image medium vessels edge: a lot of unique points obtained all are positioned at the edge of blood vessel, be unfavorable for cutting operation below, because the gray-scale value of Ink vessel transfusing pixel is starkly lower than the gray-scale value of ambient background, adjusted the position of unique point by gray value gradient, the unique point of vessel boundary is moved to Ink vessel transfusing.In S3 specifically in the following way:

S31: judge in subtraction image, whether feature point set is all processed complete, and if it is computation process terminates, if not the unique point that then taking-up one is not processed from feature point set;

S32: judge whether above-mentioned not processed unique point is the point that in 8 fields, gray-scale value is minimum, if it is continues to judge next not processed unique point;

S33: if be judged as in S32 otherwise calculate the gradient on this unique point 8 directions, using point maximum for gradient as new unique point;

S34: repeat above-mentioned steps, until feature point set is all disposed in subtraction image.

The result of unique point adjustment is as Fig. 7, Fig. 7 (a) shows the position at unique point place before adjustment, Fig. 7 (b) shows the position at unique point place after the adjustment, and the unique point being positioned at vessel boundary is substantially all positioned at the center of blood vessel after adjustment.

S4: adopt Space Rotating coordinate system model to reject the unique point of non-vascular position in sequence subtraction image: cerebrovascular image is that the two dimension of three-dimensional blood vessel structure under projection condition is expressed, spatially there is certain corresponding relation in the unique point be positioned on blood vessel, as shown in Figure 8, set up a Space Rotating coordinate system model, rotation center is O, for the some p of spatially, the projection in x, y plane is p (x ₁, y ₁), p ₁the distance of distance axis is x ₁, after have rotated certain angle [alpha], become p', the projection in x, y plane is p ₂(x ₂, y ₂), p ₂the distance of distance axis is x ₂, p ₁with p ₂relation be just carry out displacement on the horizontal scale, ordinate does not change.Utilize the positional information of unique point on adjacent sequence subtraction image, obtain the theoretical position of unique point in other subtraction image, according to the position of unique point in other subtraction images, take the mode of voting to reject the unique point of non-vascular position.In S4 specifically in the following way:

S41:S41: choose a unique point P from current subtraction image ₁(x ₁, y ₁), calculate and P in the adjacent subtraction image in current subtraction image ₁meet the unique point P of following relation ₂(x ₂, y ₂):

|x ₂-x ₁|≤4

|y ₂-y ₁|≤2

S42: according to P ₁horizontal ordinate x ₁, P ₂horizontal ordinate x ₂and P ₁and P ₂between angle [alpha]=2 ° calculate unique point P ₁distance r to rotation central axis:

$r=\frac{\sqrt{{x_2}^2-2x_1{x}_2\cos \mathrm{\α}+{x_1}^2}}{\sin \mathrm{\α}}$

S43: calculate unique point P ₁the theoretical position P projected in other subtraction images _n: set when the sequence number of the subtraction image A of pre-treatment is as i _1,, the sequence number of other subtraction images B is i _n, the angle [alpha] of the rotation between two width subtraction images is obtained by the difference of sequence number:

α＝|2×(i _n-i ₁)|；

S44: obtain unique point P according to the above-mentioned r that calculates and anglec of rotation α ₁the position p of the projection in current subtraction image B _n(x _n, y _n);

x _n＝x ₁±r×cosα

y _n＝y ₁

S45: search the unique point (x whether existing and meet following requirement from subtraction image B _m, y _m), if existed, give unique point P ₁throw a ticket, otherwise do not vote;

|x _m-x _n|≤4

|y _m-y _n|≤2

S46: setting threshold value, rejects number of votes obtained and obtains feature point set lower than after the unique point of this threshold value.Setting threshold value in practical application is 5, the feature point set obtained after rejecting error characteristic point as shown in Figure 9, Fig. 9 (a) shows the feature point set before rejecting, Fig. 9 (b) shows the feature point set after rejecting, and the position of the unique point after rejecting mostly is positioned on blood vessel.

S5: the adaptivenon-uniform sampling threshold value determining serializing subtraction image: in the subtraction image obtained, background more complicated, when background grey scale change is larger, need automatically to determine different threshold values according to the coordinate position relation of pixel, because in subtraction image, the comparison in difference of top and the bottom blood vessel is obvious, therefore, need to adopt different algorithms to obtain dynamic threshold to fluctuating plate respectively, the method that fluctuating plate divides is consistent with the division methods in second step when extract minutiae, as shown in Figure 10, upper slice different with the window size of bottom sheet, have employed the algorithm of different definite thresholds.In S5 specifically in the following way:

S51: carry out gaussian filtering operation to subtraction image, removes noise, adopts the mode as described in S22 to carry out burst process to image, the adaptivenon-uniform sampling threshold value of picture in calculating:

Centered by pixel P (x, y), get a w _up× w _upwindow, wherein w _upfor odd number;

Gray scale expectation value E in calculation window _local(x, y);

$E_{local}\left(x,y\right)=\left(\underset{i=x-w_{up}/2}{\overset{x+w_{up}/2}{\Σ}}\underset{j=y-w_{up}/2}{\overset{y+w_{up}/2}{\Σ}}P\left(i,j\right)\right)/{w_{up}}^2$

Intensity profile standard deviation sigma (x, y) in calculation window

$\mathrm{\σ}\left(x,y\right)=\sqrt{\underset{i=x-w_{up}/2}{\overset{x+w_{up}/2}{\Σ}}\underset{j=y-w_{up}/2}{\overset{y+w_{up}/2}{\Σ}}{\left(P\left(i,j\right)-E_{local}\left(x,y\right)\right)}^2}/w_{up}$

According to intensity profile standard deviation sigma (x, y) in the gray scale expectation value Elocal (x, y) in window and window, definite threshold T _up(x, y):

T _up(x,y)＝E _local(x,y)+k _up×σ(x,y)

Wherein: k _upspan is 0.1-0.3, w _upspan is 6-11;

Calculate the adaptivenon-uniform sampling threshold value of lower picture:

Calculate the integrogram of lower picture, the schematic diagram of integrogram as shown in figure 11, Figure 11 (a) is schematic diagram, Figure 11 (b) is gray-scale map, Figure 11 (c) is corresponding integral image, by formula can obtain fast in Rd region pixel value a little and, thus obtain w _down× w _downaverage gray in window, in integrogram, the value of any point I (x, y) equals the gray-scale value sum from the image upper left corner to all points in the rectangular area that this point is formed, I represents integral image, G represents subtraction image, wherein 0≤i≤x, 0≤j≤y.

I(x,y)＝sum(G(i,j))

Centered by pixel P (x, y), get a w _down× w _downwindow, wherein w _downfor odd number,

The mean value m (x, y) of the pixel in window is obtained, wherein x-w by integrogram _down/ 2≤i≤x+w _down/ 2, y-w _down/ 2≤j≤y+w _down/ 2.

$\begin{array}{c}Sum\left(G\left(i,j\right)\right)=I\left(x-\frac{w_{down}}{2},y-\frac{w_{down}}{2}\right)+I\left(x+\frac{w_{down}}{2},y+\frac{w_{down}}{2}\right)-\\ \left(I\left(x+\frac{w_{down}}{2},y-\frac{w_{down}}{2}\right)+I\left(x-\frac{w_{down}}{2},y+\frac{w_{down}}{2}\right)\right)\end{array}$

m(x,y)＝Sum(G(i,j))/w _down ²

According to the average gray value m (x, y) in window, definite threshold T _down(x, y):

T _down(x,y)＝k _down×m(x,y)

Wherein: k _downspan is 0.7-0.9, w _downspan is 45-61.

S6: the blood-vessel image segmentation based on region growing and adaptive threshold: using the unique point that obtains in S4 as Seed Points, using the adaptivenon-uniform sampling threshold value that obtains in the S5 standard as Growing law, algorithm of region growing sequence of partitions subtraction image is adopted to obtain pure cerebrovascular image, the schematic diagram of area growth process as shown in figure 12, describes the process of algorithm of region growing.Specifically in the following way:

S61: adopt in S4 and obtain the initial seed point of unique point as algorithm of region growing, create an assistant images identical with subtraction image size, color is set to white, this Seed Points not processed mistake when the color of this seed point location is white on assistant images;

S62: the Seed Points concentrating taking-up one untreated from Seed Points, the color of this seed point location is set to black by assistant images, travels through the pixel in 8 neighborhoods of this Seed Points;

S63: the color obtaining the pixel correspondence position in 8 neighborhoods of current seed point from assistant images, if the color of the pixel in neighborhood on assistant images is white, then continue the pixel in the next neighborhood of process, if the color of the pixel in neighborhood on assistant images is black, then compare the size of the gray-scale value of this pixel and the adaptivenon-uniform sampling threshold value calculated in previous step of current seed point, if the gray-scale value of this pixel is less than adaptivenon-uniform sampling threshold value, then this pixel is joined Seed Points to concentrate, continue next pixel in process neighborhood,

S64: repeat above-mentioned steps, add to come in until no longer include new Seed Points;

S65: a newly-built image identical with subtraction image size, is mapped in newly-built image the pure cerebrovascular image after obtaining segmentation by the gray-scale value of the position of Seed Points corresponding in seed point set and subtraction image.As shown in figure 13, Figure 13 (a) is flapper image, Figure 13 (b) be segmentation after blood-vessel image, can see that blood vessel is well out divided, particularly capillary, and segmentation after blood-vessel image in noise little.The present invention is based on SIFT feature, by rejecting the unique point on cerebrovascular image, unique point is all dropped on blood vessel, then the algorithm of region growing of adaptive threshold is adopted, blood vessel is split from the image of complexity, facilitate doctor to observe, and lay the first stone for later blood vessel 3 D reconstructing.This method auto Segmentation, simple to operate, do not need manual operation, and the lower time less therefore needing to spend of the algorithm complex adopted, this method robustness is comparatively strong, can provide blood-vessel image accurately for clinical operation.

The above; be only the present invention's preferably embodiment; but protection scope of the present invention is not limited thereto; anyly be familiar with those skilled in the art in the technical scope that the present invention discloses; be equal to according to technical scheme of the present invention and inventive concept thereof and replace or change, all should be encompassed within protection scope of the present invention.

Claims (7)

1., based on a cerebrovascular image partition method for multi-angle sequenced images space characteristics point set, it is characterized in that: comprise the following steps:

S1: registration is carried out to cerebrovascular flapper image and frisket image: determine geometric properties point on flapper image and frisket image, find out the feature point pairs of coupling, by the relation between the feature point pairs of coupling, process acquisition serializing subtraction image is carried out to frisket image;

S2: the geometric properties point set of abstraction sequence subtraction image;

S3: local location adjustment is carried out to the unique point at serializing subtraction image medium vessels edge: the position being adjusted unique point by gray value gradient, the unique point of vessel boundary is moved to Ink vessel transfusing;

S4: adopt Space Rotating coordinate system model to reject the unique point of non-vascular position in subtraction image: to set up a Space Rotating coordinate system model, utilize the positional information of unique point on adjacent subtraction image, obtain the theoretical position of unique point in other subtraction image, according to the position of unique point in other subtraction images, the mode of voting is taked to reject the unique point of non-vascular position;

S5: the adaptivenon-uniform sampling threshold value determining serializing subtraction image;

S6: the blood-vessel image based on region growing and adaptive threshold is split: the unique point obtained in S4 is done

For Seed Points, using the adaptivenon-uniform sampling threshold value that obtains in the S5 standard as Growing law, adopt region raw

Long algorithm sequence of partitions subtraction image obtains pure cerebrovascular image.

2. a kind of cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set according to claim 1, is further characterized in that: in S1 specifically in the following way:

S11: utilize SURF algorithm to determine geometric properties point respectively on corresponding flapper image and frisket image, calculate the proper vector of unique point;

S12: utilize FLANN adaptation to carry out proper vector coupling, obtain some matching double points;

S13: to the point matched to carrying out screening eliminating error coupling, calculate the minor increment between matching double points, the point that a reservation distance is less than the minor increment three times between matching double points is right;

S14: adopt RANSAC algorithm to remove error matching points pair, obtain the transformation matrix between two width images, obtain the frisket image after registration by transformation matrix, the frisket image after flapper image and registration is directly subtracted shadow, obtains serializing subtraction image.

3. a kind of cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set according to claim 1, is further characterized in that: in S2 specifically in the following way:

S21: utilize SIFT algorithm to extract geometric properties point set on serializing subtraction image He on frisket image;

S22: carry out upper and lower Fragmentation when obtaining the unique point of subtraction image to image, get 1/3rd of image as upper picture, all the other are as lower picture, the operation upper picture being carried out strengthening to contrast obtains more unique point;

S23: calculate the Euclidean distance between unique point in frisket image and the unique point in subtraction image, if the Euclidean distance between the unique point in the unique point in subtraction image and frisket image is less than 50, then by this unique point, it is rejected.

4. a kind of cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set according to claim 1: in S3 specifically in the following way:

S31: judge in subtraction image, whether feature point set is all processed complete, and if it is computation process terminates, if not the unique point that then taking-up one is not processed from feature point set;

S32: judge whether above-mentioned not processed unique point is the point that in 8 fields, gray-scale value is minimum, if it is continues to judge next not processed unique point;

S33: if be judged as in S32 otherwise calculate the gradient on this unique point 8 directions, using point maximum for gradient as new unique point;

S34: repeat above-mentioned steps, until feature point set is all disposed in subtraction image.

5. a kind of cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set according to claim 1: in S4 specifically in the following way:

S41: choose a unique point P from current subtraction image ₁(x ₁, y ₁), calculate and P in the adjacent subtraction image in current subtraction image ₁meet the unique point P of following relation ₂(x ₂, y ₂):

|x ₂-x ₁|≤4

|y ₂-y ₁|≤2

S42: according to P ₁horizontal ordinate x ₁, P ₂horizontal ordinate x ₂and P ₁and P ₂between angle [alpha]=2 ° calculate unique point P ₁distance r to rotation central axis:

$r=\frac{\sqrt{{x_2}^2-2x_1{x}_2\cos \mathrm{\α}+{x_1}^2}}{\sin \mathrm{\α}}$

S43: calculate unique point P ₁the theoretical position P projected in other subtraction images _n: set when the sequence number of the subtraction image A of pre-treatment is as i _1,, the sequence number of other subtraction images B is i _n, the angle [alpha] of the rotation between two width subtraction images is obtained by the difference of sequence number:

α＝|2×(i _n-i ₁)|；

S44: obtain unique point P according to the above-mentioned r that calculates and anglec of rotation α ₁the position p of the projection in current subtraction image B _n(x _n, y _n);

x _n＝x ₁±r×cosα

y _n＝y ₁

S45: search the unique point (x whether existing and meet following requirement from subtraction image B _m, y _m), if existed, give unique point P ₁throw a ticket, otherwise do not vote;

|x _m-x _n|≤4

|y _m-y _n|≤2

S46: setting threshold value, rejects number of votes obtained and obtains feature point set lower than after the unique point of this threshold value.

6. a kind of cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set according to claim 1, is further characterized in that: in S5 specifically in the following way:

S51: carry out gaussian filtering operation to subtraction image, removes noise, adopts the mode as described in S22 to carry out burst process to image, the adaptivenon-uniform sampling threshold value of picture in calculating:

Centered by pixel P (x, y), get a w _up× w _upwindow, wherein w _upfor odd number;

Gray scale expectation value E in calculation window _local(x, y);

$E_{local}\left(x,y\right)=\left(\underset{i=x-w_{up}/2}{\overset{x+w_{up}/2}{\Σ}}\underset{j=y-w_{up}/2}{\overset{y+w_{up}/2}{\Σ}}P\left(i,j\right)\right)/{w_{up}}^2$

Intensity profile standard deviation sigma (x, y) in calculation window

$\mathrm{\σ}\left(x,y\right)=\sqrt{\underset{i=x-w_{up}/2}{\overset{x+w_{up}/2}{\Σ}}\underset{j=y-w_{up}/2}{\overset{y+w_{up}/2}{\Σ}}{\left(P\left(i,j\right)-E_{local}\left(x,y\right)\right)}^2}/w_{up}$

According to intensity profile standard deviation sigma (x, y) in the gray scale expectation value Elocal (x, y) in window and window, definite threshold T _up(x, y):

T _up(x,y)＝E _local(x,y)+k _up×σ(x,y)

Wherein: k _upspan is 0.1-0.3, w _upspan is 6-11;

Calculate the adaptivenon-uniform sampling threshold value of lower picture:

Calculate the integrogram of lower picture, in integrogram, the value of any point I (x, y) equals the gray-scale value sum from the image upper left corner to all points in the rectangular area that this point is formed, I represents integral image, G represents subtraction image, wherein 0≤i≤x, 0≤j≤y;

I(x,y)＝sum(G(i,j))

Centered by pixel P (x, y), get a w _down× w _downwindow, wherein w _downfor odd number,

The mean value m (x, y) of the pixel in window is obtained, wherein x-w by integrogram _down/ 2≤i≤x+w _down/ 2, y-w _down/ 2≤j≤y+w _down/ 2;

$\begin{array}{c}\mathrm{Sum}\left(G(i,j)\right)=I_{(x-\frac{w_{\mathrm{down}}}{2},y-\frac{w_{\mathrm{down}}}{2})}+I_{(x+\frac{w_{\mathrm{down}}}{2},y+\frac{w_{\mathrm{down}}}{2})}-\\ (I_{(x+\frac{w_{\mathrm{down}}}{2},y-\frac{w_{\mathrm{down}}}{2})}+I_{(x-\frac{w_{\mathrm{down}}}{2},y+\frac{w_{\mathrm{down}}}{2})})\end{array}$

m(x,y)＝Sum(G(i,j))/w _down ²

According to the average gray value m (x, y) in window, definite threshold T _down(x, y):

T _down(x,y)＝k _down×m(x,y)

Wherein: k _downspan is 0.7-0.9, w _downspan is 45-61.

7. a kind of cerebrovascular image partition method based on multi-angle sequenced images space characteristics point set according to claim 1, is further characterized in that: in S6 specifically in the following way:

S61: adopt in S4 and obtain the initial seed point of unique point as algorithm of region growing, create an assistant images identical with subtraction image size, color is set to white, this Seed Points not processed mistake when the color of this seed point location is white on assistant images;

S62: the Seed Points concentrating taking-up one untreated from Seed Points, the color of this seed point location is set to black by assistant images, travels through the pixel in 8 neighborhoods of this Seed Points;

S63: the color obtaining the pixel correspondence position in 8 neighborhoods of current seed point from assistant images, if the color of the pixel in neighborhood on assistant images is white, then continue the pixel in the next neighborhood of process, if the color of the pixel in neighborhood on assistant images is black, then compare the size of the gray-scale value of this pixel and the adaptivenon-uniform sampling threshold value calculated in previous step of current seed point, if the gray-scale value of this pixel is less than adaptivenon-uniform sampling threshold value, then this pixel is joined Seed Points to concentrate, continue next pixel in process neighborhood,

S64: repeat above-mentioned steps, add to come in until no longer include new Seed Points;

S65: a newly-built image identical with subtraction image size, is mapped in newly-built image the pure cerebrovascular image after obtaining segmentation by the gray-scale value of the position of Seed Points corresponding in seed point set and subtraction image.