CN101853386A - Topological tree based extraction method of image texture element features of local shape mode - Google Patents

Abstract

The invention relates to the technical field of image processing, in particular to an extraction method of image texture element feature a local shape mode based on a topological tree. The method comprises the steps of: carrying out Level Set layering on an image I according to a pixel gray value v, structuring a topological tree structure T, structuring an encoded concentric circle template, scaling node shapes to be equivalent to the size of the concentric circle template, overlapping the node shapes subjected to scaling with the concentric circle template and carrying out binary coding on the node shapes subjected to scaling with the overlapping relationship of each sector fnm, counting a frequency column diagram based on coded values of all the node shapes in all the M sectors of each circle and then splicing the frequency column diagrams of the N circles, and adding all the texture feature descriptions of the shapes subjected to coding in the image topological tree. The invention can avoid the loss of texture information in the process of filtering and transformation, carry out description on the image texture more comprehensively and completely, improve the accuracy of image processing applications based on the texture element features, such as retrieval, classification, partitioning and the like,.

Description

Image texture primitive feature extracting method based on the local shape pattern of topological tree

Technical field

The present invention relates to technical field of image processing, relate in particular to a kind of image texture primitive feature extracting method of the local shape pattern based on topological tree.

Background technology

In computer vision and Flame Image Process, image texture analyses is a basic problem, yet, up to the present, people but do not form unified understanding to the explication of texture, it is generally acknowledged, texture is gradation of image or color variation or repetition spatially.Intensity profile generally has certain regularity in the texture image, and for random grain, it also has the feature on some statistical significances.People have following common recognition to texture at present:

A) texture shows as the repetition constantly in the bigger zone of this sequence of certain local sequentiality;

B) texture exists the basic comprising unit that causes visually-perceptible, i.e. texture primitive;

C) texture can not be processed into a point process, shows as region characteristic more;

D) the texture region various piece roughly is uniform entity, and each several part has roughly the same size;

E) texture has features such as intensity, density, direction and degree of roughness.

Based on above common recognition, think that texture has two key elements: (1) texture has the elementary cell that causes visually-perceptible, i.e. texture primitive.The form of texture primitive is various, shows as some image color or grayscale mode.(2) texture primitive has certain queueing discipline, and these rules may show as certain regularity, also may show as randomness (referring to document 1).

Texture analysis is one of main contents of texture research, it also is important field of research in the machine vision, boundless application background is arranged, and its application comprises that Flame Image Process (Image Processing), artificial intelligence (ArtificialIntelligence), remote Sensing Image Analysis (Remotely-sensed Image Analysis), medical image analysis (Medical ImageAnalysis), industrial surface detect (Industrial Surface Inspection), document process fields such as (Document Processing).

A key problem of texture analysis is texture description (Texture Description), is texture feature extraction (Texture Feature Extraction) at area of pattern recognition.Many texture characteristic extracting methods have been arranged at present.Tuceryan and Jain roughly are classified as four big classes with these methods: structure analysis method, statistical analysis technique, modelling analytical approach and signal processing method.Wherein statistical analysis technique and signal processing method are being served as very important role in texture analysis.

Texture analysis research at home mainly is the concrete application of a certain method.For the method for statistics, the co-occurrence matrix method is more commonly used.In the method based on model, fractal method is used many, adopts Fractal Brown function mostly, and there also have pair fractal method to carry out to be improved; The application of Markov random field (MRF) also has, and main difficulty is determining of parameter.In the method for mathematic(al) manipulation, commonly based on method of wavelet.External mainly is that textural characteristics that several texture analysis methods are extracted is in conjunction with the general classification method, to the comparison of classifying of different images.

Early stage texture analysis uses the method for statistics or structure to extract feature.Nearest major progress is to use the textural characteristics of multiresolution (for example Gabor conversion) and hyperchannel (several wave band binding analysis textural characteristics) to describe.Past lacks texture analysis effectively to be analyzed the texture of different scale, and the different scale here is meant same width of cloth image, carries out texture analysis on different scale.Can obtain the different texture feature of the same area like this, increase quantity of information, finally can improve the precision of classification.

Texture description based on filtering can be divided into two kinds of methods: horizontally-spliced one-tenth histogram after the filtering, as vertically conspiring to create a vector after GIST method (referring to document 2), the filtering, add up again, be called texture primitive method (list of references 3), texture primitive is meant micromechanism basic in the natural image, it is the fundamental element of visually-perceptible starting stage (noting the stage in advance).The research of texture primitive is to utilize the thought of sparse coding (sparse coding) to attempt the super complete image-based shading reason of study primitive s from natural image, expresses texture image with texture primitive again.

Be different from based on texture description methods such as filtering, statistics and structures image is described through filtering or certain transformation results, the present invention on the basis that the image topological tree is expressed directly to texture based on being described, can avoid the loss of texture information in the intermediate description process like this, describe thereby obtain textural characteristics more effectively, comprehensively.

Document 1: Liu Xiaomin, the texture Review Study. computer utility research, 2008, Vol.25, No.8

Document 1:Aude Oliva, " Gist of the Scene ", Neurobiology of Attention 2005

Document 3:Manik Varma, Andrew Zisserman, " A Statistical Approach to Texture Classificationfrom Single Images ", Kluwer Academic Publishers.Printed in the Netherlands, 2004

Summary of the invention

At the technical matters of above-mentioned existence, the purpose of this invention is to provide a kind of image texture primitive feature extracting method of the local shape pattern based on topological tree, to extract the feature of texture image efficiently.

For achieving the above object, the present invention adopts following technical scheme:

1. according to grey scale pixel value image is carried out the level set layering;

2. on the level set basis, make up the topological tree structure;

3. make up coding concentric circles template;

4. choose in the topological tree structure all nodes or part of nodes and carry out follow-up coding, wherein part of nodes can be the minimum shape node that comprises image pixel;

When 5. encoding, the node shape is zoomed in the concentric circles template size suitable, when the center of gravity of node shape overlapped with the concentric circles center of circle, the concentric circles border of shape and radius maximum was tangent;

When 6. encoding, node shape behind the convergent-divergent and concentric circles template is overlapping, node shape center of gravity is overlapped with the concentric circles center of circle, carry out binary coding according to the overlapping relation of node shape and each sector;

7. with each node shape frequency histogram of encoded radio statistics in all sectors of each circle, with the frequency histogram splicing of individual circle, the textural characteristics that obtains each shape is described again;

8. the textural characteristics of the shape of all participation codings in the image topological tree is described addition, get image texture features to the end.

The described level set layering of step in 1. comprises high level collection layering and low-level collection layering;

Described high level collection is layered as image according to grey scale pixel value v 〉=v ₀Be 0, v＜v ₀Be that 1 rule is converted to one group of bianry image, wherein v ₀=0,1 ..., V _MaxV is the gray-scale value of image pixel, satisfies 0≤v≤V _Max, maximal value is V _Max, for general optical imagery V _Max=255;

Described low-level collection is layered as and uses image pixel gray-scale value v≤v ₀Be 0, v＞v ₀Be that 1 rule is one group of bianry image, wherein v with image transitions ₀=V _Max... 1,0.

With in the level set every layer be that 1 shape S extracts;

Respectively high level collection and low-level concentrate, in levels, comprise or the involved relation tree that connects according to shape S;

Threaded tree with low-level collection is the topological tree structure of main design of graphics picture;

Behind the empty node in the low-level collection threaded tree of polishing, low-level collection threaded tree just becomes the topological tree of image.

The statistics of the binary coding of step described in 6., the step frequency histogram described in 7. is divided into:

Contour encoding, hard statistics with histogram; Contour encoding, soft statistics with histogram; Regional code, hard statistics with histogram; Regional code, soft statistics with histogram.

Extract the boundary profile of topological tree node shape, mate with the coding templet circle;

When the profile of node shape drops on n concentrically ringed m sector f _NmInternal clock is designated as 1, otherwise is 0; Wherein, n is concentrically ringed sequence number, satisfies 1≤n≤N, and N is concentric circles lattice numbers, N 〉=1; M is the sequence number of sector region, satisfies 1≤m≤M, and M is the number of sector region in each concentric circles, M 〉=2;

After by that analogy 0,1 mark being carried out in concentrically ringed all sectors of template, the concentrically ringed M binary marks of n is converted to decimal system numerical value;

With all participate in the encoded radio statistics frequency histogram of the shape node of coding in the topological tree, get textural characteristics histogram to the end.

Extract the boundary profile of topological tree node shape, mate with the coding templet circle;

When the profile of node shape drops on n concentrically ringed m sector f _NmWhen interior length was l, this sector mark was P _Nm=l/L, wherein L is the girth of node shape profile, if there is not the profile of node shape in the sector, then is labeled as 0;

After by that analogy mark being carried out in concentrically ringed all sectors of template, the concentrically ringed M binary marks of n is carried out once soft statistics with histogram:

$h_{\mathrm{ni}}=\underset{m=0}{\overset{M}{\mathrm{\Π}}}{b}_m\left(i\right)P_{\mathrm{nm}}+(1-b_m\left(i\right))(1-P_{\mathrm{nm}})$

B wherein _m(i) value (being 0 or 1) of m position in the binary expression of expression i.And with N concentrically ringed soft histogram splicing;

h＝{h ₁，h ₂，…，h _n，…h _N}

Wherein:

Soft histogram with the shape of all participation codings in the topological tree adds up at last:

$H=\underset{s\∈T}{\mathrm{\Σ}}h$

H is last image texture features histogram.

Zone and coding templet circle that extraction topological tree node shape comprises mate;

When n concentrically ringed m sector f dropped in the zone of node shape _NmInternal clock is designated as 1, otherwise is 0;

After by that analogy 0,1 mark being carried out in concentrically ringed all sectors of template, the concentrically ringed M binary marks of n is converted to decimal system numerical value;

With all participate in the encoded radio statistics frequency histogram of the shape node of coding in the topological tree, get textural characteristics histogram to the end.

Zone and coding templet circle that extraction topological tree node shape comprises mate;

When n concentrically ringed m sector f dropped in the zone of node shape _NmInternal clock is designated as P _Nm=s _Nm'/s _Nm, s wherein _Nm' for dropping on sector f _NmThe area of interior shape area, s _NmBe the area of sector, otherwise be labeled as 0;

After by that analogy mark being carried out in concentrically ringed all sectors of template, the concentrically ringed M binary marks of n is carried out once soft statistics with histogram:

$h_{\mathrm{ni}}=\underset{m=0}{\overset{M}{\mathrm{\Π}}}{b}_m\left(i\right)P_{\mathrm{nm}}+(1-b_m\left(i\right))(1-P_{\mathrm{nm}})$

B wherein _m(i) value (being 0 or 1) of m position in the binary expression of expression i.And with the concentrically ringed soft histogram splicing of N:

h＝{h ₁，h ₂，…，h _n，…h _N}

Wherein:

At last the soft histogram of the shape of all participation codings in the topological tree is added up and promptly gets image texture characteristic histogram to the end:

$H=\underset{s\∈T}{\mathrm{\Σ}}h$

The present invention has the following advantages and good effect:

1) by description that texture primitive is directly encoded, texture description different from the past is described based on certain filtering or the transformation results to image, can avoid texture information in filtering or conversion process, to lose, intactly image texture is described more comprehensively;

2) can change graphical rule, affine variation and rotation change have robustness preferably, the accuracy rate that Flame Image Process such as can improve retrieval based on the texture primitive feature, classify, cut apart is used.

Description of drawings

Fig. 1 is the process flow diagram of the image texture primitive feature extracting method of the local shape pattern based on topological tree of the present invention.

Fig. 2 obtains the synoptic diagram that the image topological tree is expressed among the present invention.

Fig. 3 is the synoptic diagram that makes up the coding templet circle among the present invention.

Fig. 4 A is the hard statistics with histogram synoptic diagram of contour encoding among the present invention.

Fig. 4 B is the soft statistics with histogram synoptic diagram of contour encoding among the present invention.

Fig. 4 C is the hard statistics with histogram synoptic diagram of regional code among the present invention.

Fig. 4 D is the soft statistics with histogram synoptic diagram of regional code among the present invention.

Embodiment

The image texture primitive feature extracting method based on the local shape pattern of topological tree that the present invention proposes specifically may further comprise the steps, and describes each step in detail below in conjunction with accompanying drawing 1:

Step 1, image I is carried out level set (Level Set) layering according to grey scale pixel value v:

High level collection (Upper Level Set), with image according to grey scale pixel value v 〉=v ₀Be 0, v＜v ₀Be that 1 rule is converted to one group of bianry image, wherein v ₀=0,1 ..., V _MaxV is the gray-scale value of image pixel, satisfies 0≤v≤V _Max, maximal value is V _Max, for general optical imagery V _Max=255;

In like manner, low-level collection (Lower Level Set) layering is to use image pixel gray-scale value v≤v ₀Be 0, v＞v ₀Be that 1 rule is one group of bianry image, wherein v with image transitions ₀=V _Max... 1,0; For general image, V _MaxBe 255, the two-value topological diagram picture of last low-level collection and high level collection respectively has V _Max+ 1 layer.

Step 2, on the level set basis, make up topological tree structure T:

Every layer is defined as L according to gray threshold in the level set _vThe outer contour shape S that in every layer is 1 zone is extracted, as the leaf node in the topological tree, simultaneously according to the relation of the position in image, comprise or involved relation with the shape composition in the upper and lower level set layer, in tree construction, be presented as father and son's node relationships, express as topological tree referring to Fig. 2 design of graphics, the gray-scale value of numeral 0,1,2 presentation videos wherein can be divided into following step:

1. in the level set every layer be that 1 shape S extracts, for example (corresponding shape A～G) is represented in lattice type zone respectively among the figure with reference to the shape A among the figure 2～G;

2. in levels, be contained in the involved relation tree that connects at high level collection and low-level concentrating respectively according to shape S.For example among Fig. 2, high level collection layering shape A comprises shape B, and B comprises C and D, and the threaded tree that obtains is that node A is the father node of Node B, and simultaneously, node C, D are the child nodes of Node B; In like manner, low-levelly concentrate that node F is arranged is that the child node of node E is the father node of node G again.

3. the threaded tree with low-level collection is the topological tree structure of main design of graphics picture.At first the cavity that low-level collection is connected in all nodes of seeds uses high level to collect the node polishing of threaded tree correspondence; Secondly descendants's node of polishing node is transplanted to the low-level collection threaded tree from high level collection threaded tree;

4. behind the empty node in the low-level collection threaded tree of polishing, low-level collection threaded tree just becomes the topological tree of image.For example last topological tree node E is the father node of F, D merge node among Fig. 2, and D, G node are again the child node of F, D merge node simultaneously.

Step 3, structure coding concentric circles template:

Be illustrated in figure 3 as the synoptic diagram that makes up the coding templet circle.Shown in Fig. 3 (a), setting one group of radius is r _n, { r ₁＜r ₂＜...＜r _NConcentric circles, concentric circles is divided into some sector regions according to angle θ, n concentrically ringed m sector definition is regional f _Nm

Shown in Fig. 3 (b) is the concentric circles that is made of two circles, θ=45 ° wherein, and concentric circles is divided into 8 sectors.

Step 4, choose in the topological tree structure all nodes or part of nodes carries out follow-up coding, wherein part of nodes can be the minimum shape node that comprises image pixel.

In step 5, when coding, at first zoom to the node shape in the concentric circles template size quite, and promptly when the center of gravity of node shape overlapped with the concentric circles center of circle, the concentric circles border of shape and radius maximum was tangent.

When step 6, coding, node shape behind the convergent-divergent and concentric circles template is overlapping, node shape center of gravity is overlapped, according to node shape and each sector f with the concentric circles center of circle _NmOverlapping relation carry out binary coding.

Step 7, with each node shape frequency histogram of encoded radio statistics in all M sector of each circle, the frequency histogram with N circle splices again, promptly obtains the textural characteristics description of each shape.

Step 8 is described addition with the textural characteristics of the shape of all participation codings in the image topological tree, gets image texture features to the end, referring to Fig. 4 A-4D shape coding and statistics with histogram.

In one embodiment of the present of invention, the statistics of binary coding and frequency histogram has following several mode in the step 6,7:

1. contour encoding, hard statistics with histogram is referring to Fig. 4 A

Extract the boundary profile of topological tree node shape, mate, when the profile of node shape drops on n concentrically ringed m sector f with the coding templet circle _NmInternal clock is designated as 1, otherwise is 0, by that analogy 0,1 mark is carried out in concentrically ringed all sectors of template after, the concentrically ringed M binary marks of n is converted to decimal system numerical value, for example (10101010) ₂=(170) ₁₀Deng, at last with all participate in the encoded radio statistics frequency histogram of the shape node of coding in the topological tree, get textural characteristics histogram to the end;

N is concentrically ringed sequence number, satisfies 1≤n≤N, and N is concentric circles lattice numbers, N 〉=1; M is the sequence number of sector region, satisfies 1≤m≤M, and M is the number of sector region in each concentric circles, M 〉=2, and the reference position of n, m sequence number is unimportant, as long as guarantee identical for shape n, the m reference position of all codings.

2. contour encoding, soft statistics with histogram is referring to Fig. 4 B

Extract the boundary profile of topological tree node shape, mate, when the profile of node shape drops on n concentrically ringed m sector f with the coding templet circle _NmWhen interior length was l, this sector mark was P _Nm=l/L, wherein L is the girth of node shape profile, if do not have the profile of node shape in the sector, then is labeled as 0, by that analogy mark is carried out in concentrically ringed all sectors of template after, the concentrically ringed M binary marks of n is carried out once soft statistics with histogram:

$h_{\mathrm{ni}}=\underset{m=0}{\overset{M}{\mathrm{\Π}}}{b}_m\left(i\right)P_{\mathrm{nm}}+(1-b_m\left(i\right))(1-P_{\mathrm{nm}})$

B wherein _m(i) value (being 0 or 1) of m position in the binary expression of expression i.And with the concentrically ringed soft histogram splicing of N:

h＝{h ₁，h ₂，…，h _n，…h _N}

Wherein:

Soft histogram with the shape of all participation codings in the topological tree adds up at last:

$H=\underset{s\∈T}{\mathrm{\Σ}}h$

H is last image texture features histogram;

3. regional code, hard statistics with histogram is referring to Fig. 4 C

Zone and coding templet circle that extraction topological tree node shape comprises mate, when n concentrically ringed m sector f dropped in the zone of node shape _NmInternal clock is designated as 1, otherwise is 0, by that analogy 0,1 mark is carried out in concentrically ringed all sectors of template after, the concentrically ringed M binary marks of n is converted to decimal system numerical value, for example (10101010) ₂=(170) ₁₀Deng, at last with all participate in the encoded radio statistics frequency histogram of the shape node of coding in the topological tree, get textural characteristics histogram to the end;

4. regional code, soft statistics with histogram is referring to Fig. 4 D

Zone and coding templet circle that extraction topological tree node shape comprises mate, when n concentrically ringed m sector f dropped in the zone of node shape _NmInternal clock is designated as P _Nm=s _Nm'/s _Nm, s wherein _Nm' for dropping on sector f _NmThe area of interior shape area, s _NmBe the area of sector, otherwise be labeled as 0, by that analogy mark is carried out in concentrically ringed all sectors of template after, the concentrically ringed M binary marks of n is carried out once soft statistics with histogram:

$h_{\mathrm{ni}}=\underset{m=0}{\overset{M}{\mathrm{\Π}}}{b}_m\left(i\right)P_{\mathrm{nm}}+(1-b_m\left(i\right))(1-P_{\mathrm{nm}})$

B wherein _m(i) value (being 0 or 1) of m position in the binary expression of expression i.And with the concentrically ringed soft histogram splicing of N:

h＝{h ₁，h ₂，…，h _n，…h _N}

Wherein:

At last the soft histogram of the shape of all participation codings in the topological tree is added up and promptly gets image texture characteristic histogram to the end:

$H=\underset{s\∈T}{\mathrm{\Σ}}h$

Claims (8)

1. the image texture primitive feature extracting method based on the local shape pattern of topological tree is characterized in that, may further comprise the steps:

1. according to grey scale pixel value image is carried out the level set layering;

2. on the level set basis, make up the topological tree structure;

3. make up coding concentric circles template;

4. choose in the topological tree structure all nodes or part of nodes and carry out follow-up coding, wherein part of nodes can be the minimum shape node that comprises image pixel;

5. before the coding, the node shape is zoomed in the concentric circles template size suitable, when the center of gravity of node shape overlapped with the concentric circles center of circle, the concentric circles border of shape and radius maximum was tangent;

When 6. encoding, node shape behind the convergent-divergent and concentric circles template is overlapping, node shape center of gravity is overlapped with the concentric circles center of circle, carry out binary coding according to the overlapping relation of node shape and each sector;

7. with each node shape frequency histogram of encoded radio statistics in all sectors of each circle, with the frequency histogram splicing of individual circle, the textural characteristics that obtains each shape is described again;

8. the textural characteristics of the shape of all participation codings in the image topological tree is described addition, get image texture features to the end.

2. the image texture primitive feature extracting method of the local shape pattern based on topological tree according to claim 1 is characterized in that:

The described level set layering of step in 1. comprises high level collection layering and low-level collection layering;

Described high level collection is layered as image according to grey scale pixel value v 〉=v ₀Be 0, v＜v ₀Be that 1 rule is converted to one group of bianry image, wherein v ₀=0,1 ..., V _MaxWherein, v is the gray-scale value of image pixel, satisfies 0≤v≤V _Max, maximal value is V _Max, for general optical imagery V _Max=255;

Described low-level collection is layered as and uses image pixel gray-scale value v≤v ₀Be 0, v＞v ₀Be that 1 rule is one group of bianry image, wherein v with image transitions ₀=V _Max... 1,0.

3. the image texture primitive feature extracting method of the local shape pattern based on topological tree according to claim 1 and 2 is characterized in that 2. described step further comprises following substep:

With in the level set every layer be that 1 shape S extracts;

Respectively high level collection and low-level concentrate, in levels, comprise or the involved relation tree that connects according to shape S;

Threaded tree with low-level collection is the topological tree structure of main design of graphics picture;

Behind the empty node in the low-level collection threaded tree of polishing, low-level collection threaded tree just becomes the topological tree of image.

4. the image texture primitive feature extracting method of the local shape pattern based on topological tree according to claim 1 and 2,

It is characterized in that:

The statistics of the binary coding of step described in 6., the step frequency histogram described in 7. is divided into:

Contour encoding, hard statistics with histogram; Contour encoding, soft statistics with histogram; Regional code, hard statistics with histogram; Regional code, soft statistics with histogram.

5. the image texture primitive feature extracting method of the local shape pattern based on topological tree according to claim 4 is characterized in that, described contour encoding, and hard statistics with histogram comprises following substep:

Extract the boundary profile of topological tree node shape, mate with the coding templet circle;

When the profile of node shape drops on n concentrically ringed m sector f _NmInternal clock is designated as 1, otherwise is 0, and wherein, n is concentrically ringed sequence number, satisfies 1≤n≤N, and N is concentric circles lattice numbers, N 〉=1, and m is the sequence number of sector region, satisfies 1≤m≤M, M is the number of sector region in each concentric circles, M 〉=2;

After by that analogy 0,1 mark being carried out in concentrically ringed all sectors of template, the concentrically ringed M binary marks of n is converted to decimal system numerical value;

With all participate in the encoded radio statistics frequency histogram of the shape node of coding in the topological tree, get textural characteristics histogram to the end.

6. the image texture primitive feature extracting method of the local shape pattern based on topological tree according to claim 4 is characterized in that, described contour encoding, and soft statistics with histogram comprises following substep:

Extract the boundary profile of topological tree node shape, mate with the coding templet circle;

When the profile of node shape drops on n concentrically ringed m sector f _NmWhen interior length was l, this sector mark was P _Nm=l/L, wherein L is the girth of node shape profile, if there is not the profile of node shape in the sector, then is labeled as 0;

After by that analogy mark being carried out in concentrically ringed all sectors of template, the concentrically ringed M binary marks of n is carried out once soft statistics with histogram:

$h_{\mathrm{ni}}=\underset{m=0}{\overset{M}{\mathrm{\Π}}}{b}_m\left(i\right)P_{\mathrm{nm}}+(1-b_m\left(i\right))(1-P_{\mathrm{nm}})$

B wherein _m(i) value (being 0 or 1) of m position in the binary expression of expression i.And with N concentrically ringed soft histogram splicing;

h＝{h ₁，h ₂，...，h _n，...h _N}

Wherein:

Soft histogram with the shape of all participation codings in the topological tree adds up at last:

$H=\underset{s\∈T}{\mathrm{\Σ}}h$

H is last image texture features histogram.

7. the image texture primitive feature extracting method of the local shape pattern based on topological tree according to claim 4 is characterized in that, regional code, and hard statistics with histogram comprises following substep:

Zone and coding templet circle that extraction topological tree node shape comprises mate;

When n concentrically ringed m sector f dropped in the zone of node shape _NmInternal clock is designated as 1, otherwise is 0;

After by that analogy 0,1 mark being carried out in concentrically ringed all sectors of template, the concentrically ringed M binary marks of n is converted to decimal system numerical value;

With all participate in the encoded radio statistics frequency histogram of the shape node of coding in the topological tree, get textural characteristics histogram to the end.

8. the image texture primitive feature extracting method of the local shape pattern based on topological tree according to claim 4 is characterized in that, regional code, and soft statistics with histogram comprises following substep:

Zone and coding templet circle that extraction topological tree node shape comprises mate;

When n concentrically ringed m sector f dropped in the zone of node shape _NmInternal clock is designated as

Wherein

For dropping on sector f _NmThe area of interior shape area, s _NmBe the area of sector, otherwise be labeled as 0;

After by that analogy mark being carried out in concentrically ringed all sectors of template, the concentrically ringed M binary marks of n is carried out once soft statistics with histogram:

$h_{\mathrm{ni}}=\underset{m=0}{\overset{M}{\mathrm{\Π}}}{b}_m\left(i\right)P_{\mathrm{nm}}+(1-b_m\left(i\right))(1-P_{\mathrm{nm}})$

B wherein _m(i) value (being 0 or 1) of m position in the binary expression of expression i.And with the concentrically ringed soft histogram splicing of N:

h＝{h ₁，h ₂，...，h _n，...h _N}

Wherein:

At last the soft histogram of the shape of all participation codings in the topological tree is added up and promptly gets image texture characteristic histogram to the end:

$H=\underset{s\∈T}{\mathrm{\Σ}}h.$

Images