CN1279490C - Fast morphology corrusion expansion method in image treatment - Google Patents

Abstract

The present invention relates to a fast morphology corrosion and expansion method in image processing, which belongs to the technical field of image processing. The present invention is characterized in that two sliding windows, namely a vertical sliding window and a horizontal sliding window, are added to a rectangular structure element; in the process of using the rectangular structure element to carrying out corrosion and expansion operation by moving on the whole image, the horizontal sliding window and the vertical sliding window record compared local information so as to be directly extracted and used in later calculation to realize the goal of reducing and eliminating the calculation redundancy of repeated comparison. The present invention greatly saves operation time and provides an experimental result compared with the prior art.

Description

Quick morphological erosion in the Flame Image Process, expanding method

Technical field:

The present invention relates to technical field of image processing, relate in particular to quick morphological erosion, expansion field in the Flame Image Process, holding concurrently reaches quick morphological erosion, the expanding method technology that Video processing is used.

Background technology:

Morphology is used for geology and field of ecology at first, and in recent years, some scholars proposed its thought is used for the bi-level digital image, after develop into again in the processing of gray level image and video.Now, mathematical morphology has been widely used in image noise reduction, pattern-recognition, artificial intelligence, Flame Image Process and object video the field such as cut apart, bring into play huge function.

Morphological erosion and expansion are the fundamental operations of mathematical morphology, and these two constitutes the basis of many computings after this, as open and close, restructing operation etc.Yet morphology is the burn into dilation operation that utilizes zone method relatively to define, and its method complexity and convolution are basic identical.In the process of structural element motion, produced many overlapping regions, the signaling point in the overlapping region will carry out multiple comparison, has wasted operation time, and efficient is very low.

This a part of purpose is the particularly definition of burn into expansion of brief review morphology fundamental operation, and analyzes its inefficiencies as method.

The definition of several morphology operations

Make E ⁿRepresent n dimension Euclidean space, Γ represents E ⁿThe function set of → E.

Definition 1: make B  E ⁿ, the transposition of definition B is:

$\hat{B}=\{-b|b\∈B\}---\left(1\right)$

Definition 2: make f ∈ Γ, x ∈ E ⁿ, B  E ⁿ, definition to signal f (x) with the erosion operation that point set B carries out is:

ε _B(f)(x)＝min{f(x+k)，k∈B} (2)

Title B is structural element (SE);

Definition 3: make f ∈ Γ, x ∈ E ⁿ, B  E ⁿ, definition to signal f (x) with the dilation operation that structural element B carries out is:

${\mathrm{\δ}}_B\left(f\right)\left(x\right)=\max \{f(x+k),k\∈\hat{B}\}---\left(3\right)$

Definition 4:f ∈ Γ, x ∈ E ⁿ, B  E ⁿ, definition is respectively with the open and close computing that structural element B carries out signal f (x):

γ _B(f)＝δ _B(ε _B(f)) (4)

 _B(f)＝ε _B(δ _B(f)) (5)

Above-mentioned definition be fundamental operation in the morphology, much other morphology operations are based on this several basic definitions and derive from, and repeat no more here.

From above-mentioned definition, as can be seen,, the initial point of structural element need be placed on A point place, and in the point that structural element covered, obtain minimum value, and assignment is to the A point for asking the corrosion number of some some A.Figure 1 shows that the position of structural element when in rest image, calculating A pitch corrosion value.For dilation operation, the initial point of structural element should be placed on this point, and, arrive the A point in overlay area maximizing assignment again its transposition.

Notice,, the initial point of structural element is placed on other each points that the A point can not guarantee that structural element covers all within the effective coverage of signal for the A point of signal edge.As shown in Figure 2: will produce four Null Spots that drop on outside the image border.

For this kind situation, following several disposal route can be arranged:

A. ignore Null Spot, only in the available point of structural element, try to achieve minimum value and assignment.Among Fig. 2, there are 5 data points will participate in relatively;

B. Null Spot zero setting.For the dilation operation of general gray level image, the result of this way is identical with a.But erosion operation for gray level image, may cause marginal point all to become black, so can Null Spot be set to different value (erosion operation be changed to maximum gradation value and dilation operation is changed to zero) to be unlikely to influence the continuity of edge pixel at computation performance;

C. ignore all marginal point A, like this, image whenever carries out a burn into dilation operation will be littler than original image.

Because it is few to produce the pixel of this edge effect, so adopt wherein any disposal route all not have the influence of particular importance for the processing of entire image.In this article, intend adopting second method to handle edge pixel.

More than definition itself has provided a kind of method that realizes morphological erosion, expansion.But the method that is based on definition has very big calculating redundancy.This can be as can be seen from Figure 3.

Fig. 3 (a) is the distribution situation of the initial point of structural element each point that structural element covers when moving to the A point, and the corrosion number that A is ordered should be light grey each point minimum value that covers in (a), and (b) figure is the distribution of structural element each point when moving right a pixel.The corrosion number that B is ordered is the minimum value that Dark grey covers each point.Relatively (a) and (b) two figure can find because moving of structural element produced the overlapping region, shown in (c) figure shadow region.In twice comparison procedure of the corrosion number of asking A, B, naming a person for a particular job in the overlapping region repeated comparison, causes and calculates redundancy.

In addition, in carrying out image and Video processing, what often use is the structural element of rectangle.The rectangular configuration element is simple to operate, and basic morphological operation can both be finished.At above characteristics, introduce the notion of moving window, can realize fast carrying out the morphology operations that burn into expands with the rectangular configuration element.This method can also expand in the processing of n dimensional signal.This expansion has for video sequence and important effect, because video sequence can be regarded the signal of n=3 as.In addition, will see also that the combinative class convolution algorithm (as mean filter etc.) for a certain class utilizes the thought of moving window also can construct efficiently, method fast.

Summary of the invention:

The object of the present invention is to provide a kind of is quick morphological erosion, the expanding method that the Flame Image Process of moving window notion is used based on the rectangular configuration element that slides, and can expand to the 3 D video processing, more can expand to other the combinative class convolution algorithm that on signal source, carries out by the rectangle moving window.

The invention is characterized in:

1, we have added moving window on the rectangular configuration element, and it contains next coming in order are come carries out image processing by computing machine each step:

(1) image file that will handle is read in the internal memory, deposit the width W of image in _I, height H _I, image effective coverage I, I={ (x, y) | 0≤x＜W _I, 0≤y＜H _I, and the pixel value of image each point;

(2) in computing machine, carry out the initialization setting of following parameters:

The width W of rectangular configuration element _SE, also represent the width of horizontal slip window X, its value is represented with the pixel count that this window covered, and uses X[w] (w=0,1, Λ, W _SE-1) expression horizontal slip window sequence number is the unit of w;

The height H of rectangular configuration element _SE, also represent the height of vertically sliding window mouth Y, its value is represented with the pixel count that this window covered, and uses Y _i[v] (v=0,1, Λ, H _SE-1) represents the unit that this vertical moving window sequence number is v;

The initial point O of optional rectangular configuration element, O _XBe the horizontal ordinate of this initial point present position in the rectangular configuration element, O _YOrdinate for this initial point present position in the rectangular configuration element;

It is P that image i is listed as the capable pixel of j _Ix[i] [j], wherein P _IxIt is the English symbol abbreviation of pixel;

(3) utilize the rectangular configuration element to carry out burn into dilation operation sequence of operation and moving window original state

When utilizing the rectangular configuration element to carry out erosion operation, initial point with the rectangular configuration element aligns with each pixel of image successively, the scanning entire image, scanning sequency can be any, the general selection from top to bottom, order from left to right, handle each pixel in the image successively, earlier selected enter from the upper left corner after, scanning from top to bottom again, make the initial point of structural element overlap with corresponding summit, the upper left corner in the image, this point is first pixel that requires needed erosion operation, pixel in the image that this moment, this rectangular configuration element was covered is the valid pixel value, in erosion operation, the Null Spot that drops on outside the image border all is changed to maximum gradation value MXG;

Dilation operation and erosion operation are similar, and different is wants earlier the rectangular configuration element is carried out transposition, utilize the rectangular configuration element scan image behind the transposition then, and in dilation operation, the Null Spot that drops on outside the image border all are changed to minimum gradation value MNG;

For rectangular configuration element W arbitrarily _SE, H _SE, O _X, O _Y, we are each pixel being carried out in the burn into dilation operation processing procedure, with the initial point O of structural element _X, O _YAlign with the pixel of handling, the interim coordinate system of image is set simultaneously, its initial point is the current pixel that carries out the burn into dilation operation, can obtain:

All vertical moving window Y _iOriginal state be: Y _i ^(-HSE),

The original state of all horizontal slip window X is: X _-HSE ^(-WSE),

(4) the vertical moving window Y of first of the capable j at above-mentioned first its place of pixel of initialization promptly carries out following formula and to each Y unit of vertically sliding window mouth carrying out the data filling, with near the local message the document image initial point:

$\left\{\begin{array}{cc}{Y}_i^{\left(j\right)}\left[v\right]=\mathrm{Pix}\left[i\right][j+v],& v=H_{\mathrm{SE}}-1\\ Y_i^{\left(j\right)}\left[v\right]=\min \left\{Y_i^{\left(j\right)}\right[v+1],\mathrm{Pix}[i\left]\right[j+v\left]\right\},& 0\&le;v<H_{\mathrm{SE}}-1\end{array}\right.$

Simultaneously, Y _i ^(j)[0] storage is the minimum value of all effective pixels in the vertical window in, promptly

$Y_i^{\left(j\right)}\left[0\right]=\min \left\{\mathrm{Pix}\right[i\left]\right[y\left]\right|j\&le;y<j+H_{\mathrm{SE}},(i,y)\&Element;I\};$

(5) first horizontal slip window X of the capable j at above-mentioned first its place of pixel of initialization promptly carries out following formula and to each X unit of horizontal slip window carrying out the data filling, with near the local message the document image initial point:

$\left\{\begin{array}{cc}{X}_j^{\left(i\right)}\left[w\right]=Y_{i+w}^{\left(j\right)}\left[0\right],& w=W_{\mathrm{SE}}-1\\ X_j^{\left(i\right)}\left[w\right]=\min \left\{X_j^{\left(i\right)}\right[w+1],Y_{i+w}^{\left(j\right)}[0\left]\right\},& 0\&le;w<W_{\mathrm{SE}}-1\end{array}\right.$

At this moment, X _j ⁽ⁱ⁾[0] value is to be listed as i+W with i _SEThe Y[0 of the effectively vertically moving window that-1 row link] of minimum in the value, that is:

$X_j^{\left(i\right)}\left[0\right]=\min \left\{\mathrm{Pix}\right[x\left]\right[y\left]\right|i\&le;x<i+W_{\mathrm{SE}},j\&le;x<j+H_{\mathrm{SE}},(x,y)\&Element;I\};$

(6) begin to revise pixel value from capable first pixel of this j;

(7) upgrade the horizontal slip window:

If horizontal slip window X is in state X _j ⁽ⁱ⁾, window move right a pixel and carry out renewal process after, then X becomes state X _j ⁽ⁱ⁺¹⁾

W=1 wherein, 2, Λ, W _SE-1;

$X_j^{(i+1)}[W_{\mathrm{SE}}-1]=Y_{i+W_{\mathrm{SE}}}^{\left(j\right)}\left[0\right];$

This moment X _j ⁽ⁱ⁺¹⁾[0] storage is the minimum gradation value of the pixel in the window that covers after structural element moves horizontally in;

(8) revise the rectangular configuration element and move right after the pixel this pixel value;

(9) repeat (7)～(8), all dispose until all pixels of this row;

(10) upgrade vertical moving window:

If certain vertical moving window Y _iBe in state Y _i ^(j), after the renewal, become state Y _i ^(j+1), upgrade by following formula:

V=1 wherein, 2, Λ, H _SE-1;

$Y_i^{(j+1)}[H_{\mathrm{SE}}-1]=\mathrm{Pix}\left[i\right][j+H_{\mathrm{SE}}];$

(11) repeated execution of steps (5)～(10), to the last till the delegation, this moment, entire image disposed.

In this part, will compare to the method for utilizing moving window with based on the method for definition, provide result images and based on the experimental result of other compound operations of burn into dilation operation, and set forth the expanded application of this method.

Running environment is: Pentium III 850MHz, 1G RAM.

For quick burn into dilation operation result, selection standard 8-bit gray level image, and select for use the square structure element to select following method respectively for use:

A. based on the quick erosion operation (FE) of moving window;

B. based on the erosion operation (EBD) that defines;

C. based on the rapid expanding computing (FD) of moving window;

D. based on the dilation operation (DBD) that defines.

The results are shown in Table 1 (original image: Lena:512 * 512) operation time, relatively figure line such as Figure 10.

Table 1 gray level image burn into dilation operation time (unit: ms)

For quick open and close operation result, be compared as follows method: based on the moving window burn into expand compound and the open and close computing (FO, FC), based on the burn into of definition expand compound and the open and close computing (OBD, CBD).The result is as follows:

Table 2 gray level image open and close (unit: ms) operation time

Description of drawings:

Fig. 1: calculate the corrosion number that A is ordered.

Fig. 2: the Null Spot that structural element covers.

Fig. 3: the overlapping region in the structural element motion process:

(a) initial point of structural element is at the A point;

(b) initial point of structural element moves to the B point;

(c) since structural element motion the overlapping region of overlapping generation.

Fig. 4: the vertically foundation of moving window:

(a) vertically the low sequence number unit of moving window drops on the image outside, and the numeral in is a pixel value on every side, down together;

(b) vertically moving window drops on image inside fully;

(c) vertically the high sequence number unit of moving window drops on the image outside.

Fig. 5: the foundation of horizontal slip window:

(a) the low sequence number unit of horizontal slip window drops on the image outside;

(b) the horizontal slip window drops on image inside fully;

(c) the high sequence number unit of horizontal slip window drops on the image outside.

Fig. 6: the program flow chart of the method for the invention:

(a) based on the FB(flow block) of image (or video) signal Processing of moving window method;

(b) based on the quick morphological erosion of moving window, the FB(flow block) of expanding method.

Fig. 7: the initialization of window:

(a) W _SE=5, H _SE=4, O _X=3, O _Y=2 structural element;

(b) formation of window initialization and image origin corrosion number.

Fig. 8: the renewal of horizontal slip window.

Fig. 9: the vertically renewal of moving window.

Figure 10: gray level image burn into dilation operation time comparison diagram.

Figure 11: gray level image open and close comparison diagram operation time.

Figure 12: gray level image mean filter comparison diagram operation time.

Figure 13: the actual result figure that utilization 9 * 9 structural elements carry out following 5 kinds of computings:

(a) original image;

(b) erosion operation;

(c) dilation operation;

(d) opening operation;

(e) closed operation;

(f) mean filter computing.

The preceding 15 frame erosion operation results (preceding 15 frames) of Figure 14: video sequence Foreman (352 * 288 * 300), structural element size: 5 * 5 * 5.

Figure 15: video sequence erosion operation time comparison diagram.

Embodiment:

As previously mentioned, in mathematical morphology, widespread usage be the rectangular configuration element, and initial point is in the rectangular area, that is: satisfy for structural element M: O ∈ M, O are E ⁿInitial point.This is following precondition with the moving window method that proposes.

Because the formation of rectangular configuration element is very orderly, and the overlapping region that produces in the moving process also is rectangle, and this specific character is fit to realize the moving window method.

To be example with static 8-bit gray level image below, definition, formation and the operation of moving window will be described.For the three-dimensional and even signal of higher-dimension more, be easy to the method is expanded.

Size and position with following symbolic representation image and structural element.

W _I---the width of image;

H _I---the height of image;

The effective coverage of I---image, that is: I={ (x, y) | 0≤x＜W _I, 0≤y＜H _I;

W _SE---the width of rectangular configuration element;

H _SE---the height of rectangular configuration element;

O _X---the horizontal ordinate of structural element initial point present position in the rectangular configuration element;

O _Y---the ordinate of structural element initial point present position in the rectangular configuration element;

MXG---maximum pixel gray-scale value is 255 for the 8-bit gray level image;

MNG---minimum pixel gray-scale value is 0 for the 8-bit gray level image;

Pix[i] [j]---image i is listed as the capable pixel of j.

Moving window is to be attached to the row of image and the minibuffer district that lists, and each buffer zone comprises some unit of storing a pixel value.For above-mentioned image being carried out erosion operation, need list additional H at each with the known structure element _SEThe vertical moving window of individual unit is called the moving window that links with these row.Except this W _IBeyond the individual vertical moving window, also need a W _SEThe horizontal slip window of individual unit.This horizontal slip window links with certain delegation in needs.

With with Y _iExpression and i (i=0,1, Λ, W _I-1) is listed as the vertical moving window that links, makes Y _i[v] (v=0,1, Λ, H _SE-1) represents the unit that this vertical moving window sequence number is v;

The horizontal slip window is represented with X, uses X[w] (w=0,1, Λ, W _SE-1) expression horizontal slip window sequence number is the w unit.

With vertical moving window Y _i" placement " drops on that continuous effective pixel has H at most on the vertical direction in the window when i lists _SEIndividual.Each unit of window is stored to be exactly the local message that is provided by these pixels.And can produce following three kinds of relative position relations when placing vertical moving window:

A. there is some low sequence numbers unit to drop on the image outside;

B. window is fully in image inside;

C. there is some high sequence numbers unit to drop on the image outside.

The (a) and (b) of Fig. 4, (c) have represented these several situations (H wherein respectively _SE=4).

Fig. 4 has provided the process of setting up of the data of each unit of vertical moving window equally, and is as described below:

Note Y _i ^(j)For with vertical moving window Y _iBe placed into i row, the state of No. 0 unit and j row alignment, it has covered pixel Pix[i] [j]～Pix[i] [j+H _SE-1].If (pixel is not in image inside, with virtual its gray-scale value of MXG.)

So:

$\left\{\begin{array}{cc}{Y}_i^{\left(j\right)}\left[v\right]=\mathrm{Pix}\left[i\right][j+v],& v=H_{\mathrm{SE}}-1\\ Y_i^{\left(j\right)}\left[v\right]=\min \left\{Y_i^{\left(j\right)}\right[v+1],\mathrm{Pix}[i\left]\right[j+v\left]\right\},& 0\&le;v<H_{\mathrm{SE}}-1\end{array}\right.---\left(6\right)$

Like this, Y _i ^(j)[0] storage is the minimum value of all effective pixels in the vertical window in.

That is:

$Y_i^{\left(j\right)}\left[0\right]=\min \left\{\mathrm{Pix}\right[i\left]\right[y\left]\right|j\&le;y<j+H_{\mathrm{SE}},(i,y)\&Element;I\}---\left(7\right)$

And Y _i ^(j)[v], on behalf of other local messages, these information, v ≠ 0 item will be used to upgrade window with the Y after obtaining to slide _i ^(j)[0] value.

Similar with vertical moving window, the horizontal slip window interrelates with certain delegation, and different is that vertically moving window must guarantee that each is listed as privately owned one, respectively be listed as simultaneously and use, and horizontal window is to be used in order by multirow, so have only one to get final product.

Note X _j ⁽ⁱ⁾The state of the X that expression meets the following conditions:

A. to be placed on j capable for horizontal slip window X;

B. the i column alignment is desired in its No. 0 unit;

C. all vertical moving window Y _xBe in state Y _x ^(j)

X is with the W of maximum overlay images at this moment _SEEffective pixel that individual level is continuous.(W shown in Fig. 5 (a) and (b), (c) three kinds of situations also will appear, in the relative position of it and image _SE=5).

What the horizontal slip window was asked is the local message of picture element in the whole rectangular configuration element overlay area, the comparative result that vertical moving window had obtained above this will use.So each cell data of horizontal slip window is not directly to be made of the pixel gray scale, but from each vertical moving window, obtain.

At X _j ⁽ⁱ⁾The time, X covers pixel Pix[i] [j]～Pix[i+W _SE-1] [j].(if pixel is not in image inside, with the Y[0 of virtual its column of MXG] value.)

Then:

$\left\{\begin{array}{cc}{X}_j^{\left(i\right)}\left[w\right]=Y_{i+w}^{\left(j\right)}\left[0\right]& w=W_{\mathrm{SE}}-1\\ X_j^{\left(i\right)}\left[w\right]=\min \left\{X_j^{\left(i\right)}\right[w+1],Y_{i+w}^{\left(j\right)}[0\left]\right\},& 0\&le;w<W_{\mathrm{SE}}-1\end{array}\right.---\left(8\right)$

Following formula shows, X _j ⁽ⁱ⁾[0] value is to be listed as i+W with i _SEThe Y[0 of the effectively vertically moving window that-1 row link] of minimum in the value.By (7) formula, have following formula to set up again:

$X_j^{\left(i\right)}\left[0\right]=\min \left\{\mathrm{Pix}\right[x\left]\right[y\left]\right|i\&le;x<i+W_{\mathrm{SE}},j\&le;x<j+H_{\mathrm{SE}},(x,y)\&Element;I\}---\left(9\right)$

Be X _j ⁽ⁱ⁾[0] is the effective minimum value of pixel in the rectangular area of structural element representative.Moving window method operating process is shown in Fig. 6 (a):

● at first, image or video file that we will handle read in the internal memory.

● secondly, carry out the setting of initial parameter, for example width of rectangular configuration element, highly (if handle be that video also will be provided with depth information), the horizontal ordinate and the ordinate of structural element initial point present position in the rectangular configuration element.

● once more, utilize quick morphology operations to carry out image (video) computing based on moving window.

● last, processed images (video) shows or preserves, so that further processing etc.

Shown in Fig. 6 (b), as can be seen from the figure, topmost two operations of moving window are exactly initialization and renewal based on the subroutine of the quick morphology operations of moving window.

A. initialization

Initialization operation is set up and filled in each unit of moving window according to form, and is ready for sliding.

As previously mentioned, erosion operation is the process of minimum gradation value assignment in the point of certain pixel neighborhood being given this pixel.For the correct point set of setting up this vertex neighborhood, must determine the original state of all moving windows.

As Fig. 7, (W _SE=5, H _SE=4, O _X=3, O _Y=2), the erosion operation of image origin is in fact only got minimum value in 4 effective pixels, see that easily each vertical moving window Yi should be in state Y _i ⁽²⁾, and horizontal slip window X should be in state state X _-2 ⁽³⁾, X at this moment _-2 ⁽³⁾The minimal gray of 4 effective pixels that the structural element just that [0] is write down covers.

For structural element parameter (W arbitrarily _SE, H _SE, O _X, O _Y) be not difficult to find out all vertical moving window Y _iShould be in state Y _i ^(-HSE), horizontal slip window X should be in state X _-HSE ^(-WSE)

Initialization operation divides does the initialization of vertical moving window and the initialization of horizontal slip window, and initialized process is carried out (6), (8) two formulas exactly and the unit of each moving window carried out data fill near the local message the document image initial point.

B. upgrade

What store in the moving window is certain regional local message, and erosion operation carries out entire image, and this just requires moving window is moved, and upgrades the wherein data of data cell.

As can be seen from Figure 6, after having asked certain any corrosion number of certain row, carry out the renewal of horizontal slip window; After handling the pixel of whole delegation, carry out the renewal of all vertical moving windows.

That from the above mentioned, in the moving window erosion operation is had direct use is X _j ⁽ⁱ⁾[0] value, and the data of other unit keep for the fast updating moving window.

The renewal of horizontal slip window is discussed earlier.

If horizontal slip window X is in state X _j ⁽ⁱ⁾, window move right a pixel and carry out renewal process after, then X becomes state X _j ⁽ⁱ⁺¹⁾

Renewal process is as follows:

If all vertical moving window Y _xBe in state Y _x ^(j), No. 0 cell data of i+WSE row moving window is Y _i+ WSE (j) [0].Following (10), (11) two formulas have provided the renewal operating process, and the horizontal slip window will be in state X after upgrading _j ⁽ⁱ⁺¹⁾

$X_j^{(i+1)}[W_{\mathrm{SE}}-1]Y_{i+W_{\mathrm{SE}}}^{\left(j\right)}\left[0\right]---\left(11\right)$

This moment X _j ⁽ⁱ⁺¹⁾[0] storage is the minimum gradation value of the pixel in the window that covers after structural element moves horizontally in.

Fig. 8 is under Fig. 5 (b) situation, the move right renewal process of a pixel of horizontal slip window.

For vertical moving window, its renewal process and horizontal window much at one: establish certain vertical moving window Y _iBe in state Y _i ^(j), after the renewal, become state Y _i ^(j+1), upgrade by following formula:

V=1 wherein, 2, Λ, HSE-1

(12)

$Y_i^{(j+1)}[H_{\mathrm{SE}}-1]=\mathrm{Pix}\left[i\right][j+H_{\mathrm{SE}}]---\left(13\right)$

For the vertical moving window of Fig. 4 (b) pixel that moves up, renewal process as shown in Figure 9.

Be the expanded application of moving window method below:

A. one of expanded application: mean filter

By the characteristic of moving window as can be known, this method will not only be applicable to the burn into dilation operation, and on signal source, carrying out combinative class convolution algorithm by rectangular window (as maximal value, minimum value, addition, multiplication etc.) a class problem can effectively be used, and only needing asking the operation transform of minimum (maximum) value in the method among the IV is that corresponding computing gets final product.

As utilizing the moving window method to realize mean filter MVF, the mean filter of rectangular window B is defined as follows.

$\mathrm{MVF}\left(f\right)\left(x\right)=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}f(x+k_i)(k_i\&Element;B,n=|B\left|\right)---\left(14\right)$

The quick mean filter experimental result of using moving window to carry out following (wherein FM represents the fast method result, and the MBD representative is based on the method for definition):

MBD

45.00

100.62

216.41

333.12

475.79

645.47

Table 3 gray level image mean filter (unit: ms) operation time

Figure 13 utilizes 9 * 9 structural elements to carry out the experimental result picture of above-mentioned five kinds of computings.

B. two of expanded application: the morphology operations of video sequence

This method can also be applied in the higher-dimension Signal Processing process.Often will use morphology operations in the cutting apart, extract of object in video sequence, this computing is the three-dimensional computing of carrying out for whole sequence.The moving window method can expand in the three-dimensional computing.In the moving window method of three-dimensional, also to add the moving window on some depth directions, and make it on this direction, to slide.It below is the experimental result (preceding 15 frames) of carrying out erosion operation for the video sequence Foreman (352 * 288 * 300) of CIF form with 5 * 5 * 5 structural elements.

Operation time is as shown in the table, relatively figure line such as Figure 15.

The table 4 video sequence erosion operation time (unit: s)

Be based on the operational analysis of moving window below:

The purpose of introducing moving window is to write down the local message of relatively having crossed, and can directly read and participate in comparison in the calculating afterwards, to reach minimizing and to eliminate the calculating redundancy that repeats comparison.

Time complexity is analyzed

Execution is carried out the erosion operation of image based on the morphology fast method, and the girth of operation time and structural element is linear, is analyzed as follows.

The data comparison is carried out in following process:

A. the vertically initialization of moving window:

Each vertical window initialization needs H at the most _SE-1 secondary data relatively (is looked O _YValue and decide), common W _IIndividual vertical window needs W at the most altogether _I* (H _SE-1) inferior vertical initialization data relatively;

B. the initialization of horizontal slip window:

Every row will carry out the initialization of horizontal slip window when beginning, and needs W at the most at every turn _SE-1 secondary data relatively (is looked O _XAnd decide), be total to H _IInferior initialization needs H at the most altogether _I* (W _SE-1) the sub-level initialization data relatively;

C. the renewal of horizontal slip window:

To upgrading the horizontal slip window each time, consider following two kinds of egregious cases:

C.1 best-case

If $Y_{i+W_{\mathrm{SE}}}^{\left(j\right)}\left[0\right]<X_j^{\left(i\right)}\left[0\right],$ Because $X_j^{\left(i\right)}\left[0\right]\mathrm{\&omega;}{X}_j^{\left(i\right)}\left[w\right](w>0)$ Set up,, only need once compare, can finish the renewal of horizontal window by (10), (11) formula;

C.2 worst case

If $Y_{i+W_{\mathrm{SE}}}^{\left(j\right)}\left[0\right]\&GreaterEqual;X_j^{\left(i\right)}[W_{\mathrm{SE}}-1],$ Because have $X_j^{\left(i\right)}[W_{\mathrm{SE}}-1]\&GreaterEqual;X_j^{\left(i\right)}\left[w\right](w<W_{\mathrm{SE}}-1),$ Then must carry out all WSE-1 secondary data relatively by (10), (11) formula, could upgrade horizontal window.

The probability of supposing various situations appearance is identical, then whenever carries out a horizontal window and upgrades, and the average data number of comparisons is W _SE/ 2.The update times of each row is about W _I, be total to H _IOK, be about W so be used to upgrade the average data number of comparisons of horizontal slip window _I* H _I* W _SE/ 2.

D. the vertically slip of moving window:

Each vertical moving window upgrades and once on average needs H _SE/ 2 secondary data compare, and each window upgrades H _IInferior, be total to W _IIndividual window, the average data number of comparisons that then is used to upgrade vertical moving window is W _I* H _I* H _SE/ 2.

To sum up, two-dimentional rest image is carried out erosion operation, total data number of comparisons is about:

W _I×(H _SE-1)+H _I×(W _SE-1)+W _I×H _I×(W _SE+H _SE)/2≈W _I×H _I×(W _SE+H _SE)/2

Each pixel average out to $\frac{W_I\&times;H_I\&times;(W_{\mathrm{SE}}+H_{\mathrm{SE}})/2}{W_I\&times;H_I}=\frac{W_{\mathrm{SE}}+{\mathrm{<figure-callout\; id="2"\; label="H\; SE"\; filenames="C20041000034000184.png,C20041000034000185.png"\; state="\{\{state\}\}">H</figure-callout>}}_{\mathrm{SE}}}{2}$ Inferior.

Except that above-mentioned time overhead, in the window renewal process, also have some data move operations, as 23,27, the assignment statement in 41,45 row.In fact, upgrade horizontal window and vertical window, need to carry out W respectively _SEInferior and H _SEData move, and entire image is carried out the window data that erosion operation carried out move total about W _I* H _I* (W _SE+ H _SE) secondary data moves.Each pixel average out to W _SE+ H _SEInferior.

From above discussion as can be seen, use the moving window method known image is carried out erosion operation, will be O (W working time _SE+ H _SE).

For based on the definition method, each pixel all will with W in the neighborhood _SE* H _SE-1 pixel compares, and average number of comparisons is W _SE* H _SE-1, so be O (W its working time _SE* H _SE).

The method space complexity is analyzed:

This method needs extra storage space to deposit the data of moving window, and the comparison information that buffering is local reaches the purpose that improves arithmetic speed.As previously mentioned, need W altogether _IIndividual vertical moving window, each window needs H _SEThe individual data cell that can store a pixel gray scale; Need one W is arranged in addition _SEThe horizontal slip window of individual unit.Need extra storage overhead to be like this, altogether: W _I* H _SE+ W _SEIndividual unit.For the 8-bit gray scale image, be W _I* H _SE+ W _SEIndividual byte.Relative entire image W _I* H _IThe memory space of individual unit, these extra storage are inappreciable.

Claims (1)

1, quick morphological erosion, the expanding method in the Flame Image Process, the structural element of employing rectangle, it is characterized in that: we have added moving window on the rectangular configuration element, and it contains next coming in order are come carries out image processing by computing machine each step:

(1) image file that will handle is read in the internal memory, deposit the width W of image in _I, height H _I, image effective coverage I, I={ (x, y) | 0≤x＜W _I, 0≤y＜H _I, and the pixel value of image each point;

(2) in computing machine, carry out the initialization setting of following parameters:

The width W of rectangular configuration element _SE, also represent the width of horizontal slip window X, its value is represented with the pixel count that this window covered, and uses X[w], w=0,1, Λ, W _SE-1, expression horizontal slip window sequence number is the unit of w;

The height H of rectangular configuration element _SE, also represent the height of vertically sliding window mouth Y, its value is represented with the pixel count that this window covered, and uses Y _i[v], v=0,1, Λ, H _SE-1, represent the unit that this vertical moving window sequence number is v;

The initial point O of optional rectangular configuration element, O _xBe the horizontal ordinate of this initial point present position in the rectangular configuration element, O _YOrdinate for this initial point present position in the rectangular configuration element;

It is P that image i is listed as the capable pixel of j _Ix[i] [j], wherein P _IxIt is the English symbol abbreviation of pixel;

(3) utilize the rectangular configuration element to carry out burn into dilation operation sequence of operation and moving window original state

When utilizing the rectangular configuration element to carry out erosion operation, initial point with the rectangular configuration element aligns with each pixel of image successively, the scanning entire image, scanning sequency can be any, the general selection from top to bottom, order from left to right, handle each pixel in the image successively, earlier selected enter from the upper left corner after, scanning from top to bottom again, make the initial point of structural element overlap with corresponding summit, the upper left corner in the image, this point is first pixel that requires needed erosion operation, pixel in the image that this moment, this rectangular configuration element was covered is the valid pixel value, in erosion operation, the Null Spot that drops on outside the image border all is changed to maximum gradation value MXG;

Dilation operation and erosion operation are similar, and different is wants earlier the rectangular configuration element is carried out transposition, utilize the rectangular configuration element scan image behind the transposition then, and in dilation operation, the Null Spot that drops on outside the image border all are changed to minimum gradation value MNG;

For rectangular configuration element W arbitrarily _SE, H _SE, O _X, O _Y, we are each pixel being carried out in the burn into dilation operation processing procedure, with the initial point O of structural element _x, O _YAlign with the pixel of handling, the interim coordinate system of image is set simultaneously, its initial point is the current pixel that carries out the burn into dilation operation, can obtain:

All vertical moving window Y _iOriginal state be: Y _i ^(-HSE),

The original state of all horizontal slip window X is: X _-HSE ^(-WSE),

(4) the vertical moving window Y of first of the capable j at above-mentioned first its place of pixel of initialization promptly carries out following formula and to each Y unit of vertically sliding window mouth carrying out the data filling, with near the local message the document image initial point:

$\left\{\begin{array}{cc}{Y}_i^{\left(j\right)}\left[v\right]=\mathrm{Pix}\left[i\right][j+v],& v=H_{\mathrm{SE}}-1\\ Y_i^{\left(j\right)}\left[v\right]=\min \left\{Y_i^{\left(j\right)}\right[v+1],\mathrm{Pix}[i\left]\right[j+v\left]\right\},& 0\&le;v<H_{\mathrm{SE}}-1\end{array}\right.$

Simultaneously, Y _i ^(j)[0] storage is the minimum value of all valid pixels in the vertical window in, promptly

$Y_i^{\left(j\right)}\left[0\right]=\min \left\{\mathrm{Pix}\right[i\left]\right[y\left]\right|j\&le;y<j+H_{\mathrm{SE}},(i,y)\&Element;I\};$

(5) first horizontal slip window X of the capable j at above-mentioned first its place of pixel of initialization promptly carries out following formula and to each X unit of horizontal slip window carrying out the data filling, with near the local message the document image initial point:

$\left\{\begin{array}{cc}{X}_j^{\left(i\right)}\left[w\right]=Y_{i+w}^{\left(j\right)}\left[0\right],& w=W_{\mathrm{SE}}-1\\ X_j^{\left(i\right)}\left[w\right]=\min \left\{X_j^{\left(i\right)}\right[w+1],Y_{i+w}^{\left(j\right)}[0\left]\right\},& 0\&le;w<W_{\mathrm{SE}}-1\end{array}\right.$

At this moment, X _j ⁽ⁱ⁾[0] value is to be listed as i+W with i _SEThe Y[0 of the effectively vertically moving window that-1 row link] of minimum in the value, that is:

$X_j^{\left(i\right)}\left[0\right]=\min \left\{\mathrm{Pix}\right[x\left]\right[y\left]\right|i\&le;x<i+W_{\mathrm{SE}},j\&le;x<j+H_{\mathrm{SE}},(x,y)\&Element;I\};$

(6) begin to revise pixel value from capable first pixel of this j;

(7) upgrade the horizontal slip window:

If horizontal slip window X is in state X _j ⁽ⁱ⁾, window move right a pixel and carry out renewal process after, then X becomes state X _j ⁽ⁱ⁺¹⁾,

W=1 wherein, 2, Λ, W _SE-1;

$X_j^{(i+1)}[W_{\mathrm{SE}}-1]=Y_{i+W_{\mathrm{SE}}}^{\left(j\right)}\left[0\right];$

This moment X _j ⁽ⁱ⁺¹⁾[0] storage is the minimum gradation value of the pixel in the window that covers after structural element moves horizontally in;

(8) revise the rectangular configuration element and move right after the pixel this pixel value;

(9) repeat (7)～(8), all dispose until all pixels of this row;

(10) upgrade vertical moving window:

If certain vertical moving window Y _iBe in state Y _i ^(j), after the renewal, become state Y _i ^(j+1), upgrade by following formula:

V=1 wherein, 2, Λ, H _SE-1;

$Y_i^{(j+1)}[H_{\mathrm{SE}}-1]=\mathrm{Pix}\left[i\right][j+H_{\mathrm{SE}}];$

(11) repeated execution of steps (5)～(10), to the last till the delegation, this moment, entire image disposed.

Images