CN102663762B - The dividing method of symmetrical organ in medical image - Google Patents

Abstract

The present invention relates to Medical Image Processing, technical field of computer vision.Considering the symmetric properties of organ for providing a kind of, the dividing method of symmetrical organ in speed and precision problem medical image can be solved, for achieving the above object, the technical scheme that the present invention takes is, does the dividing method of symmetrical organ in medical image, comprise the steps: step one: the restriction frame Bounding determining image? Box; Step 2: detect axis of symmetry, calculates symmetrical mapping matrix; Step 3: obtain symmetrical similar matrix; Step 4: introduce symmetry constraint model in the drawings; Step 5: solve energy equation by figure segmentation method, obtains final segmentation result.The present invention is mainly used in Medical Image Processing.

Description

The dividing method of symmetrical organ in medical image

Technical field

The present invention relates to Medical Image Processing, technical field of computer vision, particularly relate to the dividing method of symmetrical organ in a kind of medical image.

Background technology

Digital Image Processing refers to and converts picture signal to digital signal and utilize the process that computing machine processes it.Along with the deep development of image processing techniques, the great achievement that Digital Image Processing obtains is application on the medical image.In a width medical image, people are usually only interested in partial organ wherein, and these organs occupy certain region usually, and image that is upper in some characteristic (as gray scale, profile, color and texture etc.) and that close on has difference.These characteristic differences may clearly, also may be very trickle, so that naked eyes can not be discovered out.Medical image segmentation is one of prerequisite of its identification and understanding, and the quality of Iamge Segmentation quality directly affects the effect of successive image process, even determines its success or failure.Segmentation or three-dimensional visualization, virtual operation basis.So medical image cutting method has very important effect in field of medical image processing.

Traditional medical image cutting method [1] comprises Threshold segmentation, Region growing segmentation [2], Interactive Segmentation, LiveWire segmentation, NormalizedCut segmentation [3], the segmentation of deformation model, the segmentation of fuzzy connectedness, the segmentation based on proxy machine model.On the one hand, because the uneven of medical image histoorgan and wriggling cause the uneven and fuzzy of pixel grey scale, and local bulk effect and pathological tissues cause tissue and pathology edge fog, contrast low, and these problems propose very large challenge to existing dividing method.On the other hand, tissue and the organ of human body much all have symmetry, and the low level feature that traditional medical image cutting method utilizes, as region and edge, well do not utilize the symmetric properties of image.

Patented claim US2010081931A1 proposes a kind of method of Iamge Segmentation, in an at least dimeric picture, first around object, manually demarcates several point, goes to the edge of matching object to be split with curve.According to the edge detected, picture is tentatively divided into several part, then the statistical distribution of every a part of gray-scale value is obtained, the proportion shared by every a part of gray-scale value is obtained according to statistical distribution, because statistical is furnished with multiple parameter, application expectation value maximizes (EM) algorithm and carries out parameter estimation.Finally obtain estimated parameter and weight, application Exploration/Selection algorithm is split.Patent US200913002490 proposes a kind of method of medical image segmentation, first the volume profiles index (SI) of each voxel in picture is calculated, connect SI feature, intensity and locus, form the set of eigenvectors of one five dimension, finally introduce expectation value and maximize algorithm, the information provided in conjunction with spatial-intensity carries out the segmentation of pathology or tumour picture.This patent achieves better effects in the medical image of clear-cut, but this patent does not consider the symmetric properties of organ.Patent US7336809B2 proposes convolution dividing method, for solving speed and precision problem, is divided into two stages: presegmentation and fine segmentation.In the presegmentation stage, with the convolution mark picture background of image, reduce the resolution of picture.Then the anatomical structure mark that the merogenesis stage obtains is refined and be segmented in further to fine segmentation, finally obtains the tissue regions be concerned about.This patent has good performance in splitting speed and precision, but this patent does not consider the symmetric properties of organ.

Summary of the invention

The present invention is intended to overcome the deficiencies in the prior art, there is provided a kind of and consider the symmetric properties of organ, the dividing method of symmetrical organ in speed and precision problem medical image can be solved, for achieving the above object, the technical scheme that the present invention takes is, in medical image, the dividing method of symmetrical organ, comprises the steps:

Step one: the restriction frame BoundingBox determining image

For the initial pictures provided, need user to provide one and limit frame, as initial input;

Step 2 detects axis of symmetry, calculates symmetrical mapping matrix

Application MIFT algorithm obtains discrete match point, and MIFT is the acronym of MirrorReflectionInvariantFeatureDescriptor, by the match point obtained, uses RANSAC method to calculate the mapping matrix of 2 dimensions, uses H _trepresent;

Step 3: obtain symmetrical similar matrix

Utilize symmetrical mapping matrix H _t, obtain pixel u initial matched pixel u ', by the optimizing process of local, obtain one of pixel u symmetrical coupling more accurately, similar discrimination m (k) of the neighbor point k of definition u ' is:

$m\left(k\right)={\left|\right|z_u-z_k\left|\right|}_2\·\exp \left(\frac{\mathrm{dis}(x_{u^{\′}},x_k)}{10}\right),---\left(1\right)$

Here z _urepresent the feature comprising brightness, color of pixel u, || z _u-z _k|| ₂represent that the pixel k and pixel u that detect are on feature space, dis (x _{u '}, x _k) represent pixel x _{u '}and x _keuclidean distance, the symmetrical match point v of the best of pixel u with can obtain by minimizing similar discrimination m:

$u\→v=\arg \underset{v\∈N\left(u^{\′}\right)}{\min }\left\{m\left(v\right)\right\},---\left(2\right)$

Here N (u ') is the set of the vicinity points of the symmetric points u ' of u, and definition maps m={m (u) } the similar discrimination set of representative picture, described the symmetrical structure of the overall situation by entropy, setting balance parameters λ is:

$\mathrm{\λ}=\exp \left\{\frac{\mathrm{Entrop}{y}^3\left(m\right)}{1000}\right\},---\left(3\right)$

Entropy (m) represents the entropy of similar discrimination set, and above 3 represent cube, and 1000 is normalized parameter, for adjustment factor, according to the symmetric case of picture, obtain different λ values, last by the axis of symmetry obtained and mapping matrix, the symmetrical similar matrix of picture is represented H _s, symmetrical similar matrix H _s=[H _s(u)] use exponential form definition:

$H_S\left(u\right)=2\mathrm{\λexp}(-\frac{m\left(u\right)}{\mathrm{mean}\left(m\right)})+1-\mathrm{\λ},---\left(4\right)$

Wherein mean (m) is the mean value of m, H _su () represents Comparability, H _su the value of () is limited between [1-λ, 1+ λ] by parameter lambda;

Step 4: introduce symmetry constraint model in the drawings

Here need symmetrical similar matrix to introduce markov random file, obtain the symmetric properties of foreground object:

Traditional Markov stochastic equation:

$E\left(L\right)=\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{D}_u\left(L_u\right)+\underset{(u,v)\∈N}{\mathrm{\Σ}}{V}_{u,v}(L_u,L_v),---\left(5\right)$

Here L={L _u| u ∈ Λ } represent the two-value label of picture Λ, for generally splitting L _u{ 0,1}, 0 represents background to ∈, and 1 represents organ, for pixel v, L _v∈ { 0,1}; Data item D _ube called data penalty, its acquisition is based on the prospect of image and background; V _{u, v}be level and smooth punishment, it is used for encouraging similar pixel to distribute same label; N comprises the right set of adjacent pixels, and use symmetrical similar matrix as conversion parameter to strengthen the extraction of symmetrical prospect, Update Table penalty is D ' _u:

$D_u^{\′}\left(u\right)=\left\{\begin{array}{ccc}v(u;L_u)\mathrm{Hs}\left(u\right)& \mathrm{if}& L_u=s\\ v(u;L_u)(2-\mathrm{Hs}\left(u\right))& \mathrm{if}& L_u=t\end{array}\right.,---\left(6\right)$

Wherein Hs (u) represents Comparability, and s represents source point, and t represents meeting point, at data penalty D ' _uin, encourage symmetrical pixel to be prospect, asymmetric portion markings is background;

Identical with data penalty, the symmetry of linearly connected coupling, to the level and smooth punishment with Markov model, is expressed as:

$\underset{\{u,v\}\∈N}{\mathrm{\Σ}}{V}_{u,v}(L_u,L_v)+\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{V}_{u\→v}^S\left(L_u\right),---\left(7\right)$

Here u → v represents that pixel v is the symmetric points of the pixel u obtained by equation (2), and other variable is as smoothly punished V _{u, v}the same with equation (5), after the S of a top represent the limit of u to v;

Step 5: solve energy equation by figure segmentation method, obtains final segmentation result

Finally, energy equation is defined as:

$E_s\left(L\right)=\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{D}_u^{\′}\left(L_u\right)+\underset{\{u,v\}\∈N}{\mathrm{\Σ}}{V}_{u,v}(L_u,L_v)+\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{V}_{u\→v}^S\left(L_u\right),---\left(8\right)$

Now segmentation can be regarded as global minimization, i.e. L ^*=argminE _s(L), because the energy equation E of definition in equation (8) _s(L) have submodularity, solve so the minimized problem of energy equation can cut algorithm by figure.

Restriction frame is rectangular shaped rim.

MIFT algorithmic procedure is as follows:

1) for input picture, represent with multiscale space;

2) detect yardstick spatial extrema point, screen unique point: the extreme point obtained constitutes the alternative set of unique point from extreme point, unique point screens in the alternative point in set thus, weeds out the point of low contrast and skirt response in this set;

3) precise positioning feature point position: through operating the unique point determined above, adopts the mode of the three-dimensional quafric curve of matching carry out the matching of information and approach, to obtain more accurate coordinate and graphical rule information;

4) determine direction parameter: have rotational invariance to meet unique point, assisting the information of gradient direction on neighborhood territory pixel and size according to unique point, is each unique point assigned direction parameter;

5) unique point descriptor coding.According to the original SIFT coded sequence of direction parameter dynamic conditioning, to obtain overturning constant feature descriptor MIFT;

Right by the coupling obtained, use RANSAC method to calculate the mapping matrix H of 2 dimensions _t, make Λ be the dot matrix of all pixels of input picture, for each pixel u ∈ Λ, symmetric coordinates are geometrically: x _{u '}=H _tx _u, x here _u=(x _u, y _u, 1) ^tprovide the homogeneous coordinates of each pixel u, subscript T represents transposition, x _{u '}the homogeneous coordinates according to converting the geometry symmetric position obtained.

Technical characterstic of the present invention and effect:

In the dividing method of the symmetrical organ of a kind of medical image of the present invention's proposition, by providing restriction frame, detect the axis of symmetry of picture, axis of symmetry is utilized to calculate the mapping matrix of organ in picture, then the symmetric matrix of organ is calculated, set up according to the restricted model of organ symmetric properties, by organ segmentation out, thus reach efficient and segmentation result accurately.Method of the present invention considers the symmetry of human organ, tissue, and segmentation accuracy rate has significant improvement.The present invention also can be adapted to the segmentation of the picture object of complex background, and method is simple, and intuitively, time complexity is low, and accuracy is high.

Accompanying drawing explanation

Fig. 1: general frame of the present invention.

Fig. 2: source point optimal match point schematic diagram.

Fig. 3: data penalty, symmetry smoothly punish schematic diagram.(1) data penalty; (2) symmetry is smoothly punished.

Fig. 4: input picture schematic diagram.In figure, (a) inputs picture.B () user has detected the picture of axis of symmetry after providing and limiting frame.

Fig. 5: the pictorial information form of symmetric matrix is represented result figure.(a) pictorial information after symmetry coupling.B symmetrical similar matrix information that () calculates.

Fig. 6: a few Zhang Butong there is symmetric medical picture.

Fig. 7: experimental result picture: the left side is input picture, the right is for exporting.

Embodiment

The present invention is directed to traditional medicine image partition method Problems existing, in based on the image partition method of low level, introduce high-rise symmetry constraint.Fig. 1 shows the framework of this method: first, and user provides restriction frame to input picture; Then symmetrical mapping matrix is obtained by symmetry detection algorithm; Then symmetrical similar matrix is introduced to strengthen symmetrical mapping; The last symmetry constraint of introducing in the drawings model, and by figure segmentation method segmentation picture.

Main contributions of the present invention is the symmetric properties utilizing organ, promotes the segmentation effect of organ, and the symmetry constraint model that the present invention proposes has submodularity [4], can be solved by figure segmentation method in polynomial time.

Image partition method based on axis of symmetry has five steps, as shown in figure (1).Here is the detailed description to this five steps:

Step one: the restriction frame (BoundingBox) determining image

The present invention, for the initial pictures provided, needs user to provide one and limits frame, as initial input.Because for background segment before image, need to provide an initial input to eliminate the ambiguity (namely specifying fore/background initial range) of picture prospect, background by user.Provide by user interactions the segmentation [5] that the method limiting frame is widely used in display foreground, background, the present patent application be input as a rectangular shaped rim.

Step 2 detects axis of symmetry, calculates symmetrical mapping matrix

Symmetry is the build-in attribute of a lot of organ, and in two-dimentional geometric figure, the symmetry of plane can carry out projection transform.In the present invention, MIFT[6 is applied] (AMirrorReflectionInvariantFeatureDescriptor) algorithm obtains discrete match point.MIFT algorithmic procedure is as follows:

1., for input picture, represent with multiscale space.

2. detect yardstick spatial extrema point, from extreme point, screen unique point.The extreme point obtained constitutes the alternative set of unique point, unique point by thus set in alternative point in screen.Because and the extreme point of not all meets the requirement of unique point.In this set, also there is the point of low contrast and skirt response, as their uniqueness of feature of image and stability outstanding not, so this two classes point will be weeded out.

3. precise positioning feature point position.Have passed through above operation, unique point is determined, because the impact in the conversion on yardstick and pixel unit size may cause unique point coordinate and some deviation of dimensional information, accurate in order to ensure the information of unique point, the mode of the three-dimensional quafric curve of matching is adopted to carry out the matching of information and approach, to obtain more accurate coordinate and graphical rule information.

4. determine direction parameter.Have rotational invariance to meet unique point, assisting the information of gradient direction on neighborhood territory pixel and size according to unique point, is each unique point assigned direction parameter.

5. unique point descriptor coding.According to the original SIFT coded sequence of direction parameter dynamic conditioning, to obtain overturning constant feature descriptor MIFT.

Right by the coupling obtained, use RANSAC method [7] to calculate the mapping matrix of 2 dimensions, use H here _trepresent.Make Λ be the dot matrix of all pixels of input picture, for each pixel u ∈ Λ, symmetric coordinates are geometrically: x _{u '}=H _tx _u, x here _u=(x _u, y _u, 1) ^tprovide the homogeneous coordinates of each pixel u, subscript T represents transposition, x _{u '}the homogeneous coordinates according to converting the geometry symmetric position obtained.

Step 3: obtain symmetrical similar matrix

Utilize mapping matrix H _t, pixel u can obtain its initial matched pixel u '.Third step, by the optimizing process of local, obtains one of pixel u symmetrical coupling more accurately.Should be adjacent with initial matched pixel u ' on this matching theory.Similar discrimination m (k) that we define the neighbor point k of u ' is:

$m\left(k\right)={\left|\right|z_u-z_k\left|\right|}_2\·\exp \left(\frac{\mathrm{dis}(x_{u^{\′}},x_k)}{10}\right),---\left(1\right)$

Here z _urepresent the feature (as brightness, color) of pixel u.As shown in Figure 2: || z _u-z _k|| ₂represent that the pixel k and source point u that detect are on feature space.Dis (x _{u '}, x _k) represent pixel x _{u '}and x _keuclidean distance, be to regulate the order of magnitude divided by 10.

Fig. 2: u is source point, the initial symmetric points that u ' detects for second step.Because u ' may not be optimal match point, so at dis (x _{u '}, x _k) distance in look for one with source point optimal match point k.

L2 normal form is used to the distance in controlling feature space, and exponential term dis () is the punishment of space length, makes symmetric points meet constraint on space geometry position: near the point of geometry symmetric position, to be worth larger.Finally, the symmetrical match point v of the best of pixel u with can obtain by minimizing similar discrimination m:

$u\→v=\arg \underset{v\∈N\left(u^{\′}\right)}{\min }\left\{m\left(v\right)\right\},---\left(2\right)$

Here N (u ') is the set of the vicinity points of the symmetric points u ' of u.Definition maps m={m (u) } the similar discrimination set of representative picture.By observing us find, relative to the foreground area of symmetry, the histogram of the similar discrimination of asymmetric background area is more level and smooth.Therefore can be described the symmetrical structure of the overall situation by entropy, we set balance parameters λ and are:

$\mathrm{\λ}=\exp \left\{\frac{\mathrm{Entrop}{y}^3\left(m\right)}{1000}\right\},---\left(3\right)$

Here Entropy (m) represents the entropy of similar discrimination set, and above 3 represent cube.1000 is normalized parameter, for adjustment factor.According to the symmetric case of picture, obtain different λ values.Last by the axis of symmetry obtained and mapping matrix, by picture symmetric matrix (note: be different from mapping matrix H above _t) represent, symmetric matrix H _s=[H _s(u)] use exponential form definition:

$H_S\left(u\right)=2\mathrm{\λexp}(-\frac{m\left(u\right)}{\mathrm{mean}\left(m\right)})+1-\mathrm{\λ},---\left(4\right)$

Wherein mean (m) is the mean value of m, symmetrical similar matrix H _su the codomain of () is limited between [1-λ, 1+ λ] by parameter lambda.

Step 4: introduce symmetry constraint model in the drawings

Here need symmetrical similar matrix to introduce markov random file, obtain the symmetric properties of foreground object.As shown in Figure 3, the symmetry constraint model in the present invention is made up of two parts: (1) data penalty; (2) symmetry is smoothly punished.

Fig. 3: s is source point, and t is meeting point.The thickness of line represents the punishment size of this edge, and in the drawings, some u and some v is symmetrical.A () does not consider symmetric properties, its segmenting structure may be wrong.B the red dotted line in () represents axis of symmetry, the line of the yellow introduced in (b) is that u smoothly punishes for the symmetry of v, and our method can provide correct label to u, v.

Traditional Markov stochastic equation:

$E\left(L\right)=\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{D}_u\left(L_u\right)+\underset{(u,v)\∈N}{\mathrm{\Σ}}{V}_{u,v}(L_u,L_v),---\left(5\right)$

Here L={L _u| u ∈ Λ } represent the two-value label of picture Λ.For generally splitting L _u{ 0,1}, 0 represents background to ∈, and 1 represents organ, for pixel v, L _v∈ { 0,1}; Data item D _ube called data penalty, its acquisition is based on the prospect of image and background; V _{u, v}be level and smooth punishment, it is used for encouraging similar pixel to distribute same label; N comprises the right set of adjacent pixels.In the present invention, we use symmetrical similar matrix as conversion parameter to strengthen the extraction of symmetrical prospect, and Update Table penalty is D ' _u:

$D_u^{\′}\left(u\right)=\left\{\begin{array}{ccc}v(u;L_u)\mathrm{Hs}\left(u\right)& \mathrm{if}& L_u=s\\ v(u;L_u)(2-\mathrm{Hs}\left(u\right))& \mathrm{if}& L_u=t\end{array}\right.,---\left(6\right)$

Wherein Hs (u) represents Comparability, and s represents source point, and t represents meeting point.At data penalty D ' _uin, encourage symmetrical pixel to be prospect, asymmetric part is more prone to be labeled as background.

Identical with data penalty, the symmetry of linearly connected coupling, to the level and smooth punishment with Markov model, is expressed as:

$\underset{\{u,v\}\∈N}{\mathrm{\Σ}}{V}_{u,v}(L_u,L_v)+\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{V}_{u\→v}^S\left(L_u\right),---\left(7\right)$

Here u → v represents that pixel v is the symmetric points of the pixel u obtained by equation (2).Other variable is as smoothly punished V _{u, v}the same with equation (5), after the S of a top represent the limit of u to v.

Step 5: solve energy equation by figure segmentation method, obtains final segmentation result

Finally, energy equation is defined as:

$E_s\left(L\right)=\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{D}_u^{\′}\left(L_u\right)+\underset{\{u,v\}\∈N}{\mathrm{\Σ}}{V}_{u,v}(L_u,L_v)+\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{V}_{u\→v}^S\left(L_u\right),---\left(8\right)$

Now segmentation can be regarded as global minimization, i.e. L ^*=argminE _s(L).Because the energy equation E of definition in equation (8) _s(L) have submodularity, solve so the minimized problem of energy equation can cut algorithm by figure.

In the dividing method of the symmetrical organ of a kind of medical image of the present invention's proposition, by providing restriction frame, detect the axis of symmetry of picture, axis of symmetry is utilized to calculate the mapping matrix of organ in picture, then the symmetric matrix of organ is calculated, set up according to the restricted model of organ symmetric properties, by organ segmentation out, thus reach efficient and segmentation result accurately.Our method considers the symmetry of human organ, tissue, and segmentation accuracy rate has significant improvement.The present invention also can be adapted to the segmentation of the picture object of complex background, and method is simple, and intuitively, time complexity is low, and accuracy is high.

Overall flow of the present invention is as follows:

1, input picture and provided by user and limit frame

To input picture as shown in figure (4) (a), user limits frame according to oneself wanting the prospect split to provide and calculates axis of symmetry, and as figure (4) (b), providing picture is respectively example:

2, symmetry coupling is carried out

According to the axis of symmetry set up, calculate symmetrical mapping matrix H _t.By formula (1), obtain the similar discrimination of neighbor point, symmetry coupling is carried out, as figure (5) (a) to the image at two ends, left and right.

3, symmetrical similar matrix is calculated

Here for the symmetrical method that we are described, for other symmetric case, by applying different mapping matrixes, method of the present invention is still suitable for.The axis of symmetry that utilization obtains above and mapping matrix, application of formula (4), represents the form of pictorial information with symmetric matrix, and result is as shown in figure (5) (b):

4, picture segmentation

Symmetric properties according to object sets up restricted model, and application drawing cuts algorithm by object segmentation out.By above-mentioned steps, obtain final segmentation result, figure (6) be a few Zhang Butong there is symmetric medical picture.Be that application GrabCut algorithm [8] and our symmetry division method are split respectively, picture provides the comparative result of GrabCut and method of the present invention.

Below more experimental result:

The left side is input picture, and the right is for exporting.

Conclusion: when splitting symmetrical prospect, the present invention proposes a kind of symmetry constraint model.Our symmetry constraint model very naturally combines with MRF model, and strong in symmetric case, weakly can be suitable for when even not having.Another advantage is the present invention's just application symmetry constraint reconstruction image, does not affect segmentation optimization.Can be solved in polynomial time by our restricted model of figure segmentation method.Our method is not only applicable to the image of Mirror Symmetry, can introduce this framework equally for other situations (such as: rotate).

Leading reference

[1] Tian Jie, Bao Shanglian, Zhou Mingquan. " medical image processing and analysis ", Electronic Industry Press, 2003

[2] Senthilkumar., B.Umamaheswari, G.Karthik, J. " on breast cancer detects a kind of new region growing segmentation algorithm ", computer intelligence and computing research international conference, 2010, P1-4

[3] JianboShi; Malik, J. " NormalizedCut and Iamge Segmentation ", pattern analysis and machine intelligence periodical, 2000, P888-905

[4] Iwata, S.; Nagano, K. " modular function covered under constraint minimizes ", the meeting of computer science basic science, 2009, P671-680

[5] V.Lempitsky, P.Kohli, C.Rother and T.Sharp. " image partition method based on limiting frame priori ", international computer visual conference, 2009, P277-284

[6] X, Guo., X, Cao., J, Zhang. and X, Li. " MIFT: a kind of mirror-reflection Invariance feature describes ", Asia computer vision meeting, 2009, P536-544

[7] RichardSzeliski. " computer vision: algorithm and application ", Springer publishing house, 2010

[8] C.Rother, V.Kolmogorov, A.Blake. " GrabCut: adopt the mutual foreground extracting method that iteration diagram cuts ", international graphics conference, 2004,23 (3)

Claims (3)

1. the dividing method of symmetrical organ in medical image, is characterized in that, comprise the steps:

Step one: the restriction frame BoundingBox determining image

For the initial pictures provided, need user to provide one and limit frame, as initial input;

Step 2: detect axis of symmetry, calculates symmetrical mapping matrix

Application MIFT algorithm obtains discrete match point, and MIFT is the acronym of MirrorReflectionInvariantFeatureDescriptor, by the match point obtained, uses RANSAC method to calculate the mapping matrix of 2 dimensions, uses H _trepresent;

Step 3: obtain symmetrical similar matrix

Utilize symmetrical mapping matrix H _t, obtain pixel u initial matched pixel u ', by the optimizing process of local, obtain one of pixel u symmetrical coupling more accurately, similar discrimination m (k) of the neighborhood pixels k of definition u ' is:

$m\left(k\right)={\left|\right|z_u-z_k\left|\right|}_2\·\exp \left(\frac{\mathrm{dis}(x_{u^{\′}},x_k)}{10}\right),---\left(1\right)$

Here z _urepresent the feature comprising brightness, color of pixel u, || z _u-z _k|| ₂represent that pixel k and pixel u is on feature space, dis (x _{u '}, x _k) represent the Euclidean distance of pixel u ' and k, x _{u '}, x _kbe respectively the homogeneous coordinates of pixel u ', k, the symmetrical match point v of the best of pixel u obtains by minimizing similar discrimination m:

$u\→v=\underset{v\∈N\left(u^{\′}\right)}{\arg \min \left\{m\left(v\right)\right\}},---\left(2\right)$

Here N (u ') is the set of the vicinity points of the matched pixel u ' of u, and definition maps m={m (u) } the similar discrimination set of representative picture, described the symmetrical structure of the overall situation by entropy, setting balance parameters λ is:

$\mathrm{\λ}=\exp \left\{\frac{\mathrm{Entrop}{y}^3\left(m\right)}{100}\right\},---\left(3\right)$

Entropy (m) represents the entropy of similar discrimination set, and above 3 represent cube, and 1000 is normalized parameter, for adjustment factor, according to the symmetric case of picture, obtain different λ values, last by the axis of symmetry obtained and mapping matrix, by the symmetrical similar matrix H of picture _srepresent, symmetrical similar matrix H _s=[H _s(u)] use exponential form definition:

$H_S\left(u\right)=2\mathrm{\λexp}(-\frac{m\left(u\right)}{\mathrm{mean}\left(m\right)})+1-\mathrm{\λ},---\left(4\right)$

Wherein mean (m) is the mean value of m, H _su () represents Comparability, H _su the value of () is limited between [1-λ, 1+ λ] by parameter lambda;

Step 4: introduce symmetry constraint model in the drawings

Here need symmetrical similar matrix to introduce markov random file, obtain the symmetric properties of foreground object:

Traditional Markov stochastic equation:

$E\left(L\right)=\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{D}_u\left(L_u\right)+\underset{(u,v)\∈N}{\mathrm{\Σ}}{V}_{u,v}(L_u,L_v),---\left(5\right)$

Here L={L _u| u ∈ Λ } represent the two-value label of picture Λ, for generally splitting L _u{ 0,1}, 0 represents background to ∈, and 1 represents organ, for pixel v, L _v∈ { 0,1}; Data item D _ube called data penalty, its acquisition is based on the prospect of image and background; V _u,vbe level and smooth punishment, it is used for encouraging similar pixel to distribute same label; N comprises the right set of adjacent pixels, and use symmetrical similar matrix as conversion parameter to strengthen the extraction of symmetrical prospect, Update Table penalty is D ' _u:

$D_u^{\′}\left(u\right)=\left\{\begin{array}{ccc}v(u;L_u)\mathrm{Hs}\left(u\right)& \mathrm{if}& L_u=s\\ v(u;L_u)(2-\mathrm{Hs}\left(u\right))& \mathrm{if}& L_u=t\end{array}\right.,---\left(6\right)$

Wherein Hs (u) represents Comparability, and s represents source point, and t represents meeting point, at data penalty D ' _uin, encourage symmetrical pixel to be prospect, asymmetric portion markings is background;

Identical with data penalty, the symmetry of linearly connected coupling, to the level and smooth punishment with Markov model, is expressed as:

$\underset{\{u,v\}\∈N}{\mathrm{\Σ}}{V}_{u,v}(L_u,L_v)+\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{V}_{u\→v}^S\left(L_u\right),---\left(7\right)$

Here u → v represents that pixel v is the symmetric points of the pixel u obtained by equation (2), smoothly punishes V _u,vthe same with equation (5), after the S of a top represent the limit of u to v;

Step 5: solve energy equation by figure segmentation method, obtains final segmentation result

Finally, energy equation is defined as:

$E_s\left(L\right)=\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{D}_u^{\′}\left(L_u\right)+\underset{\{u,v\}\∈N}{\mathrm{\Σ}}{V}_{u,v}(L_u,L_v)+\underset{u\∈\mathrm{\Λ}}{\mathrm{\Σ}}{V}_{u\→v}^S\left(L_u\right),---\left(8\right)$

Now segmentation is regarded as global minimization, i.e. L ^*=argminE _s(L), because the energy equation E of definition in equation (8) _s(L) have submodularity, solve so the minimized problem of energy equation cuts algorithm by figure.

2. the dividing method of symmetrical organ in medical image as claimed in claim 1, it is characterized in that, restriction frame is rectangular shaped rim.

3. the dividing method of symmetrical organ in medical image as claimed in claim 1, it is characterized in that, MIFT algorithmic procedure is as follows:

1) for input picture, represent with multiscale space;

2) detect yardstick spatial extrema point, screen unique point: the extreme point obtained constitutes the alternative set of unique point from extreme point, unique point screens in the alternative point in set thus, weeds out the point of low contrast and skirt response in this set;

3) precise positioning feature point position: through operating the unique point determined above, adopts the mode of the three-dimensional quafric curve of matching carry out the matching of information and approach, to obtain more accurate coordinate and graphical rule information;

4) determine direction parameter: have rotational invariance to meet unique point, assisting the information of gradient direction on neighborhood territory pixel and size according to unique point, is each unique point assigned direction parameter;

5) unique point descriptor coding, according to the original SIFT coded sequence of direction parameter dynamic conditioning, to obtain overturning constant feature descriptor MIFT;

By the match point obtained, RANSAC method is used to calculate the mapping matrix H of 2 dimensions _t, make Λ be the dot matrix of all pixels of input picture, for each pixel u ∈ Λ, symmetric coordinates are geometrically: x _{u '}=H _tx _u, x here _u=(x _u, y _u, 1) ^tprovide the homogeneous coordinates of each pixel u, subscript T represents transposition, x _{u '}the homogeneous coordinates according to converting the geometry symmetric position obtained.