CN103440646A - Similarity obtaining method for color distribution and texture distribution image retrieval - Google Patents

Abstract

The invention discloses a similarity obtaining method for color distribution and texture distribution image retrieval and relates to the technology of image retrieval. The similarity obtaining method is characterized in that the similarity obtaining method includes the steps of extracting the color distribution characteristics of an input image and the texture distribution characteristics of the input image, calculating the similarity between the color distribution characteristics of the input image and the color distribution characteristics of each image in a database respectively to obtain the color distribution characteristic similarity Sa (i) between the input image and each image in the database, calculating the similarity between the texture distribution characteristics of the input image and the texture distribution characteristics of each image in a database respectively to obtain the texture distribution characteristic similarity Sb (i) between the input image and each image in the database, and utilizing the formula S (i)= Wa*Sa (i) + Wb*Sb (i) to calculate the combination similarity S (i) between the input image and each image in the database.

Description

Similarity acquisition methods for color distribution and grain distribution image retrieval

Technical field

The present invention relates to image retrieval technologies, especially a kind of similarity acquisition methods for color distribution and grain distribution image retrieval.

Background technology

In recent years, along with developing rapidly of mobile Internet, the application of taking pictures has obtained very large development space, and obtain and the storage of photo becomes very easy.Along with the picture data explosive growth, the user is in the urgent need to the automatic technology of retrieval and the arrangement of comparison film.Existing image retrieval technologies all wants the training sample of prestored digital image in dependency database to try to achieve similarity.And the photo of at present cloud storage is all the photo of the various scenes of taking from various users basically, do not have the training sample that retrievable demonstration has marked.Thereby existing image retrieval technologies inconvenience is applied directly in the retrieval of cloud memory image.

Summary of the invention

Technical matters to be solved by this invention is: for the problem of above-mentioned existence, provide a kind of similarity acquisition methods that is applicable to cloud storage color distribution and grain distribution image retrieval.

Similarity acquisition methods for color distribution and grain distribution image retrieval provided by the invention, is characterized in that, comprising:

Step 1: Color Distribution Features and the grain distribution feature of extracting input picture;

Step 2: the similarity of calculating respectively the Color Distribution Features of each width image in the Color Distribution Features of described input picture and database, obtain the Color Distribution Features similarity Sa(i between each width image in input picture and database), i gets 0,1,2 ... database images sum-1;

Calculate respectively the similarity of the grain distribution feature of each width image in the grain distribution feature of described input picture and database, obtain the grain distribution characteristic similarity Sb(i between each width image in input picture and database), i gets 0,1,2 ... database images sum-1;

Step 3: utilize formula S (i)=Wa * Sa(i)+Wb * Sb(i), i gets 0,1,2 ... database images sum-1, Wa, Wb are weighting coefficient and Wa+Wb=1, the combination similarity S(i of each width image in calculating input image and database).

Preferably, the acquisition methods of described Color Distribution Features comprises:

Step 201: image transitions, to the hsv color space, is obtained to image I;

Step 202: the H of each pixel of image, S, V component are mapped as to color feature value G:G=Q _s* Q _v* H+Q _v* S+V; The span of three passages in hsv color space is carried out to interval division, be divided into respectively H _i, S _j, V _k, 0≤i≤Q wherein _h, 0≤j≤Q _s, 0≤k≤Q _v, Q _h, Q _s, Q _vthe divided interval sum of three passages that means respectively the hsv color space;

Step 203: the eigenwert distribution situation of each pixel in statistical picture: the color feature value that travels through each pixel, statistics falls into the pixel quantity of each color distribution histogram, to fall into the pixel quantity of each color distribution histogram respectively divided by image slices vegetarian refreshments sum, obtain normalized Color Distribution Features hist (x), wherein x representative color distribution histogram interval.

Preferably, the acquisition methods of described Color Distribution Features also comprises:

Image is divided into to the N piece; In described step 203: the eigenwert distribution situation of each pixel in statistical picture: travel through the eigenwert of each pixel, statistics falls into the pixel quantity of each color distribution histogram, and will not be twice of pixel in image boundary piece statistics; The pixel quantity that falls into each color distribution histogram, respectively divided by image slices vegetarian refreshments sum, is obtained to normalized Color Distribution Features hist (x), wherein x representative color histogram.

Preferably, the acquisition methods of described grain distribution feature comprises:

Step 301: be gray-scale map by image transitions, obtain image L;

Step 302: to be of a size of the template of 3 pixels * 3 pixels, travel through described image L, obtain the LBP feature of each template, the method that wherein obtains template LBP feature comprises:

The gray-scale value of 9 pixels in the note template is p _i(0≤i≤8), wherein the grey scale pixel value of template center is designated as p ₀; The gray-scale value of other pixel in template is deducted to p ₀obtain:

g _i=p _i-p ₀,(1≤i≤8)；

The g that each is calculated _icarry out binary conversion treatment: if g _i>=0 makes g _i=1, otherwise g _i=0; To be positioned at the g of the pixel of position i _ivalue expands to 82 system numbers, obtains LBP (i) and is characterized as, 1≤i≤8:

$\mathrm{LBP}\left(i\right)=\underset{q=0}{\overset{7}{\mathrm{\Σ}}}{g}_i\×2^q;$

Step 303: the LBP that obtains the invariable rotary of each template ^rifeature; Wherein obtain the LBP of the invariable rotary of template ^rithe method of feature comprises:

By carrying out shifting function, can obtain respectively 8 binary data to each LBP (i) of template, get a wherein minimum LBP as invariable rotary ^ri(i) feature:

1≤i in formula≤8, ROR means shifting function, q means the figure place that is shifted;

Step 304: the LBP that adds up each invariable rotary in each template ^ri(i) distribution situation of feature: the LBP that travels through each invariable rotary of each template ^ri(i) eigenwert, statistics falls into the pixel quantity of each grain distribution histogram, to fall into again the pixel quantity of each grain distribution histogram respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature hist (y), wherein y represents the grain distribution histogram.

Preferably, in described step 2, the method for calculating Color Distribution Features similarity Sa comprises:

Step 401: utilize formula

calculate the Color Distribution Features similarity, wherein hist ₁(x) be the Color Distribution Features of the first width image, hist ₂(x) be the Color Distribution Features of the second width image.

Preferably, in described step 2, the method for calculating grain distribution characteristic similarity Sb comprises:

Step 401: utilize formula

calculate the grain distribution characteristic similarity, wherein hist ₁(y) be the grain distribution feature of the first width image, hist ₂(y) be the grain distribution feature of the second width image.

Preferably, described Wa > Wb.

The present invention also protects a kind of acquisition methods of the similarity for the grain distribution image search method, comprising:

Step 1: the grain distribution feature of extracting input picture;

Step 2: the similarity of calculating respectively the grain distribution feature of each width image in the grain distribution feature of described input picture and database, obtain the grain distribution characteristic similarity Sb(i between each width image in input picture and database), i gets 0,1,2 ... database images sum-1;

The acquisition methods of described grain distribution feature comprises:

Step 301: be gray-scale map by image transitions, obtain image L;

Step 302: to be of a size of the template of 3 pixels * 3 pixels, travel through described image L, obtain the LBP feature of each template, the method that wherein obtains template LBP feature comprises:

The gray-scale value of 9 pixels in the note template is p _i(0≤i≤8), wherein the grey scale pixel value of template center is designated as p ₀; The gray-scale value of other pixel in template is deducted to p ₀obtain:

g _i=p _i-p ₀,(1≤i≤8)；

The g that each is calculated _icarry out binary conversion treatment: if g _i>=0 makes g _i=1, otherwise g _i=0; To be positioned at the g of the pixel of position i _ivalue expands to 82 system numbers, obtains LBP (i) and is characterized as, 1≤i≤8:

$\mathrm{LBP}\left(i\right)=\underset{q=0}{\overset{7}{\mathrm{\Σ}}}{g}_i\×2^q;$

Step 303: the LBP that obtains the invariable rotary of each template ^rifeature; Wherein obtain the LBP of the invariable rotary of template ^rithe method of feature comprises:

By carrying out shifting function, can obtain respectively 8 binary data to each LBP (i) of template, get a wherein minimum LBP as invariable rotary ^ri(i) feature:

1≤i in formula≤8, ROR means shifting function, q means the figure place that is shifted;

Step 304: the LBP that adds up each invariable rotary in each template ^ri(i) distribution situation of feature: the LBP that travels through each invariable rotary of each template ^ri(i) eigenwert, statistics falls into the pixel quantity of each grain distribution histogram, to fall into again the pixel quantity of each grain distribution histogram respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature hist (y), wherein y represents the grain distribution histogram.

The present invention also protects the acquisition methods of the grain distribution feature in above-mentioned.

In sum, owing to having adopted technique scheme, the invention has the beneficial effects as follows:

The image similarity acquisition methods the present invention relates to does not need image is carried out to any hypothesis, does not need a large amount of mark sample training models yet, has easy realization, the advantage that computing velocity is fast.

The accompanying drawing explanation

Examples of the present invention will be described by way of reference to the accompanying drawings, wherein:

Fig. 1 is that in the present invention, Color Distribution Features extracts process flow diagram.

Fig. 2 is grain distribution feature extraction process flow diagram in the present invention.

Fig. 3 is image retrieval process flow diagram in the present invention.

Embodiment

Disclosed all features in this instructions, or the step in disclosed all methods or process, except mutually exclusive feature and/or step, all can combine by any way.

Disclosed arbitrary feature in this instructions, unless special narration all can be replaced by other equivalences or the alternative features with similar purpose.That is,, unless special narration, each feature is an example in a series of equivalences or similar characteristics.

The invention provides a kind of similarity acquisition methods for color distribution and grain distribution image retrieval, its concrete steps comprise:

Step 1: Color Distribution Features and the grain distribution feature of extracting input picture;

Step 2: the similarity of calculating respectively the Color Distribution Features of each width image in the Color Distribution Features of described input picture and database, obtain the Color Distribution Features similarity Sa(i between each width image in input picture and database), i gets 0,1,2 ... database images sum-1;

Calculate respectively the similarity of the grain distribution feature of each width image in the grain distribution feature of described input picture and database, obtain the grain distribution characteristic similarity Sb(i between each width image in input picture and database), i gets 0,1,2 ... database images sum-1;

Step 3: utilize formula S (i)=Wa * Sa(i)+Wb * Sb(i), i gets 0,1,2... database images sum-1, and Wa, Wb are weighting coefficient and Wa+Wb=1, the combination similarity S(i of each width image in calculating input image and database).Because people more are concerned about color in the ordinary course of things, therefore as one preferred embodiment, weighting coefficient Wa > Wb.

As Fig. 3, after the combination similarity obtained between each width image of input picture and database, each similarity is sorted, the larger explanation two width images of similarity are more similar, we can rule of thumb set a threshold value, be greater than the image output in all databases of this threshold value by being greater than the combination similarity, as result for retrieval.

As Fig. 1, in one embodiment of the invention, the acquisition methods of Color Distribution Features comprises:

Step 201: image transitions, to the hsv color space, is obtained to image I; In general picture is the RGB color space, and it is technology well known in the art that the picture of RGB color space is transformed into to the hsv color space, does not repeat them here its detailed process.

Step 202: by the H of each pixel of image, S, V component according to formula G=Q _s* Q _v* H+Q _v* the S+V mapping relations are converted to color feature value G; Wherein, Q _h, Q _s, Q _vdefinition be such: the span of three passages in hsv color space is carried out to interval division, is divided into respectively H _i, S _j, V _k, 0≤i≤Q wherein _h, 0≤j≤Q _s, 0 ￡ k≤Q _v, Q _h, Q _s, Q _vthe divided interval sum of three passages that means respectively the hsv color space;

Step 203: the eigenwert distribution situation of each pixel in statistical picture: the color feature value that travels through each pixel, statistics falls into the pixel quantity of each color distribution histogram, to fall into the pixel quantity of each color distribution histogram respectively divided by image slices vegetarian refreshments sum, obtain normalized Color Distribution Features hist (x), wherein x representative color distribution histogram interval.

Those skilled in the art all know, and the color distribution histogram is divided into some intervals by the color characteristic of entire image, then by each pixel, in each interval situation about distributing, describe different color shared ratio in entire image.

Consider the implication that piece image is expressed, often be positioned near the zone of image boundary not too important, we more are concerned about the content that the non-borderline region of figure is expressed.Therefore, in another embodiment, the acquisition methods of described Color Distribution Features also comprises:

Image is divided into to the N piece, and for example N equals 36; The pixel be included in the image boundary piece is only added up once, and rest of pixels is added up twice.Particularly, in described step 203: the eigenwert distribution situation of each pixel in statistical picture: travel through the eigenwert of each pixel, statistics falls into the pixel quantity of each color distribution histogram; And will not the pixel statistics twice in the image boundary piece, that is, when having the eigenwert that is not the pixel in the image boundary piece to fall into a certain color distribution histogram, by falling into this interval pixel quantity, add 2; Be the eigenwert of the pixel in the image boundary piece while falling into a certain color distribution histogram when having, will fall into this interval pixel quantity and add 1; The last pixel quantity that will fall into again each color distribution histogram, respectively divided by image slices vegetarian refreshments sum, obtains normalized Color Distribution Features hist (x), wherein x representative color histogram.

The Color Distribution Features value come out like this is more accurate.

As Fig. 2, in another embodiment of the present invention, the acquisition methods of described grain distribution feature comprises:

Step 301: be gray-scale map by image transitions, obtain image L; By the RGB image transitions, be that gray-scale map has several different methods, wherein a kind of is to utilize formula L=0.299*R+0.587*G+0.114*B to be changed, wherein, and the red component of R represent pixel, the green component of G represent pixel, the blue component of B represent pixel.0.299,0.587,0.114 be coefficient, this coefficient is not unique certainly, can not be interpreted as limitation of the present invention.

Step 302: to be of a size of the template of 3 pixels * 3 pixels, travel through described image L, obtain the LBP feature (being textural characteristics) of each template, the method that wherein obtains template LBP feature comprises:

The gray-scale value of 9 pixels in the note template is p _i(0≤i≤8), wherein the grey scale pixel value of template center is designated as p ₀; The gray-scale value of other pixel in template is deducted to p ₀obtain:

g _i=p _i-p ₀,(1≤i≤8)；

The g that each is calculated _icarry out binary conversion treatment: if g _i>=0 makes g _i=1, otherwise g _i=0; The gi value that will be positioned at the pixel of position i expands to 82 system numbers, obtains LBP (i) and is characterized as, 1≤i≤8:

$\mathrm{LBP}\left(i\right)=\underset{q=0}{\overset{7}{\mathrm{\Σ}}}{g}_i\×2^q;$

The LBP feature of above-mentioned calculating can not be tackled the requirement of invariable rotary, in order to obtain the LBP feature of invariable rotary, needs further execution step 303: the LBP that obtains the invariable rotary of each template ^rifeature; Wherein obtain the LBP of the invariable rotary of template ^rithe method of feature comprises:

By carrying out shifting function, can obtain respectively 8 binary data to each LBP (i) of template, get a wherein minimum LBP as invariable rotary ^ri(i) feature:

1≤i in formula≤8, ROR means shifting function, q means the figure place that is shifted;

Step 304: the LBP that adds up each invariable rotary in each template ^ri(i) distribution situation of feature: the LBP that travels through each invariable rotary of each template ^ri(i) eigenwert, statistics falls into the pixel quantity of each grain distribution histogram, to fall into again the pixel quantity of each grain distribution histogram respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature hist (y), wherein y represents the grain distribution histogram.

Here the histogrammic definition of grain distribution and the definition of aforementioned color distribution histogram are similar.The grain distribution histogram is divided into some intervals by the textural characteristics of entire image, then by each pixel, in each interval situation about distributing, describes different texture shared ratio in entire image.

After the Color Distribution Features that obtains image, calculate the embodiment of the Color Distribution Features similarity Sa of two width images, comprising:

Step 401: utilize formula

calculate the Color Distribution Features similarity, wherein hist ₁(x) be the Color Distribution Features of the first width image, hist ₂(x) be the Color Distribution Features of the second width image, wherein x representative color histogram.

After the grain distribution feature that obtains image, calculate the embodiment of the grain distribution characteristic similarity Sb of two width images, comprising:

Step 501: utilize formula calculate the grain distribution characteristic similarity, wherein hist ₁(y) be the grain distribution feature of the first width image, hist ₂(y) be the grain distribution feature of the second width image.Y represents the grain distribution histogram.

Under the instruction of foregoing, those skilled in the art easily expect a kind of acquisition methods of the similarity for the grain distribution image search method based on innovative idea of the present invention, comprising:

Step 1: the grain distribution feature of extracting input picture;

Step 2: the similarity of calculating respectively the grain distribution feature of each width image in the grain distribution feature of described input picture and database, obtain the grain distribution characteristic similarity Sb(i between each width image in input picture and database), i gets 0,1,2 ... database images sum-1.

In like manner, after the grain distribution characteristic similarity obtained between each width image of input picture and database, each similarity is sorted, the larger explanation two width images of similarity are more similar, we can rule of thumb set a threshold value, be greater than the image output in all databases of this threshold value by being greater than the combination similarity, as result for retrieval.

The present invention is not limited to aforesaid embodiment.The present invention expands to any new feature or any new combination disclosed in this manual, and the arbitrary new method disclosed or step or any new combination of process.

Claims (10)

1. the similarity acquisition methods for color distribution and grain distribution image retrieval, is characterized in that, comprising:

Step 1: Color Distribution Features and the grain distribution feature of extracting input picture;

Step 2: the similarity of calculating respectively the Color Distribution Features of each width image in the Color Distribution Features of described input picture and database, obtain the Color Distribution Features similarity Sa(i between each width image in input picture and database), i gets 0,1,2 ... database images sum-1;

Calculate respectively the similarity of the grain distribution feature of each width image in the grain distribution feature of described input picture and database, obtain the grain distribution characteristic similarity Sb(i between each width image in input picture and database), i gets 0,1,2 ... database images sum-1;

Step 3: utilize formula S (i)=Wa * Sa(i)+Wb * Sb(i), i gets 0,1,2 ... database images sum-1, Wa, Wb are weighting coefficient and Wa+Wb=1, the combination similarity S(i of each width image in calculating input image and database).

2. the similarity acquisition methods for color distribution and grain distribution image retrieval according to claim 1, is characterized in that, the acquisition methods of described Color Distribution Features comprises:

Step 201: image transitions, to the hsv color space, is obtained to image I;

Step 202: the H of each pixel of image, S, V component are mapped as to color feature value G:G=Q _s* Q _v* H+Q _v* S+V; The span of three passages in hsv color space is carried out to interval division, be divided into respectively H _i, S _j, V _k, 0≤i≤Q wherein _h, 0≤j≤Q _s, 0≤k≤Q _v, Q _h, Q _s, Q _vthe divided interval sum of three passages that means respectively the hsv color space;

Step 203: the color feature value distribution situation of each pixel in statistical picture: the color feature value that travels through each pixel, statistics falls into the pixel quantity of each color distribution histogram, to fall into the pixel quantity of each color distribution histogram respectively divided by image slices vegetarian refreshments sum, obtain normalized Color Distribution Features hist (x), wherein x representative color distribution histogram interval.

3. the similarity acquisition methods for color distribution and grain distribution image retrieval according to claim 2, is characterized in that, the acquisition methods of described Color Distribution Features also comprises:

Image is divided into to the N piece; In described step 203: the eigenwert distribution situation of each pixel in statistical picture: travel through the eigenwert of each pixel, statistics falls into the pixel quantity of each color distribution histogram, and will not be twice of pixel in image boundary piece statistics; The pixel quantity that falls into each color distribution histogram, respectively divided by image slices vegetarian refreshments sum, is obtained to normalized Color Distribution Features hist (x), wherein x representative color histogram.

4. the similarity acquisition methods for color distribution and grain distribution image retrieval according to claim 1 and 2, is characterized in that, the acquisition methods of described grain distribution feature comprises:

Step 301: be gray-scale map by image transitions, obtain image L;

Step 302: to be of a size of the template of 3 pixels * 3 pixels, travel through described image L, obtain the LBP feature of each template, the method that wherein obtains template LBP feature comprises:

The gray-scale value of 9 pixels in the note template is p _i(0≤i≤8), wherein the grey scale pixel value of template center is designated as p ₀; The gray-scale value of other pixel in template is deducted to p ₀obtain:

g _i=p _i-p ₀,(1≤i≤8)；

The gi that each is calculated carries out binary conversion treatment: if g _i>=0 makes g _i=1, otherwise g _i=0; To be positioned at the g of the pixel of position i _ivalue expands to 82 system numbers, obtains LBP (i) and is characterized as, 1≤i≤8:

$\mathrm{LBP}\left(i\right)=\underset{q=0}{\overset{7}{\mathrm{\Σ}}}{g}_i\×2^q;$

Step 303: the LBP that obtains the invariable rotary of each template ^rifeature; Wherein obtain the LBP of the invariable rotary of template ^rithe method of feature comprises:

By carrying out shifting function, can obtain respectively 8 binary data to each LBP (i) of template, get a wherein minimum LBP as invariable rotary ^ri(i) feature:

1≤i in formula≤8, ROR means shifting function, q means the figure place that is shifted;

Step 304: the LBP that adds up each invariable rotary in each template ^ri(i) distribution situation of feature: the LBP that travels through each invariable rotary of each template ^ri(i) eigenwert, statistics falls into the pixel quantity of each grain distribution histogram, to fall into again the pixel quantity of each grain distribution histogram respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature hist (y), wherein y represents the grain distribution histogram.

5. the similarity acquisition methods for color distribution and grain distribution image retrieval according to claim 4, is characterized in that, the method for calculating Color Distribution Features similarity Sa in described step 2 comprises:

Step 401: utilize formula calculate the Color Distribution Features similarity, wherein hist ₁(x) be the Color Distribution Features of the first width image, hist ₂(x) be the Color Distribution Features of the second width image.

6. the similarity acquisition methods for color distribution and grain distribution image retrieval according to claim 4, is characterized in that, the method for calculating grain distribution characteristic similarity Sb in described step 2 comprises:

Step 401: utilize formula calculate the grain distribution characteristic similarity, wherein hist ₁(y) be the grain distribution feature of the first width image, hist ₂(y) be the grain distribution feature of the second width image.

7. the similarity acquisition methods for color distribution and grain distribution image retrieval according to claim 1, is characterized in that described Wa > Wb.

8. the acquisition methods of the similarity for the grain distribution image search method, is characterized in that, comprising:

Step 1: the grain distribution feature of extracting input picture;

Step 2: calculate respectively the similarity of the grain distribution feature of each width image in the grain distribution feature of described input picture and database, the some grain distribution characteristic similarity Sb(i that arrive), i gets 0,1,2 ... database images sum-1;

The acquisition methods of described grain distribution feature comprises:

Step 301: be gray-scale map by image transitions, obtain image L;

Step 302: to be of a size of the template of 3 pixels * 3 pixels, travel through described image L, obtain the LBP feature of each template, the method that wherein obtains template LBP feature comprises:

The gray-scale value of 9 pixels in the note template is p _i(0≤i≤8), wherein the grey scale pixel value of template center is designated as p ₀; The gray-scale value of other pixel in template is deducted to p0 to be obtained:

g _i=p _i-p ₀,(1≤i≤8)；

The g that each is calculated _icarry out binary conversion treatment: if g _i>=0 makes g _i=1, otherwise g _i=0; To be positioned at the g of the pixel of position i _ivalue expands to 82 system numbers, obtains LBP (i) and is characterized as, 1≤i≤8:

$\mathrm{LBP}\left(i\right)=\underset{q=0}{\overset{7}{\mathrm{\Σ}}}{g}_i\×2^q;$

Step 303: the LBP that obtains the invariable rotary of each template ^rifeature; Wherein obtain the LBP of the invariable rotary of template ^rithe method of feature comprises:

By carrying out shifting function, can obtain respectively 8 binary data to each LBP (i) of template, get a wherein minimum LBP as invariable rotary ^ri(i) feature:

1≤i in formula≤8, ROR means shifting function, q means the figure place that is shifted;

Step 304: the LBP that adds up each invariable rotary in each template ^ri(i) distribution situation of feature: the LBP that travels through each invariable rotary of each template ^ri(i) eigenwert, statistics falls into the pixel quantity of each grain distribution histogram, to fall into again the pixel quantity of each grain distribution histogram respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature hist (y), wherein y represents the grain distribution histogram.

9. the acquisition methods of the similarity for the grain distribution image retrieval according to claim 8, is characterized in that, the method for calculating grain distribution characteristic similarity Sb in described step 2 comprises:

Step 401: utilize formula

calculate the grain distribution characteristic similarity, wherein hist ₁(y) be the grain distribution feature of the first width image, hist ₂(y) be the grain distribution feature of the second width image.

10. the acquisition methods of a grain distribution feature, is characterized in that, comprising:

Step 301: be gray-scale map by image transitions, obtain image L;

Step 302: to be of a size of the template of 3 pixels * 3 pixels, travel through described image L, obtain the LBP feature of each template, the method that wherein obtains template LBP feature comprises:

The gray-scale value of 9 pixels in the note template is p _i(0≤i≤8), wherein the grey scale pixel value of template center is designated as p ₀; The gray-scale value of other pixel in template is deducted to p ₀obtain:

g _i=p _i-p ₀,(1≤i≤8)；

The g that each is calculated _icarry out binary conversion treatment: if g _i>=0 makes g _i=1, otherwise g _i=0; To be positioned at the g of the pixel of position i _ivalue expands to 82 system numbers, obtains LBP (i) and is characterized as, 1≤i≤8:

$\mathrm{LBP}\left(i\right)=\underset{q=0}{\overset{7}{\mathrm{\Σ}}}{g}_i\×2^q;$

Step 303: the LBP that obtains the invariable rotary of each template ^rifeature; Wherein obtain the LBP of the invariable rotary of template ^rithe method of feature comprises:

By carrying out shifting function, can obtain respectively 8 binary data to each LBP (i) of template, get a wherein minimum LBP as invariable rotary ^ri(i) feature:

1≤i in formula≤8, ROR means shifting function, q means the figure place that is shifted;

Step 304: the LBP that adds up each invariable rotary in each template ^ri(i) distribution situation of feature: the LBP that travels through each invariable rotary of each template ^ri(i) eigenwert, statistics falls into the pixel quantity of each grain distribution histogram, to fall into again the pixel quantity of each grain distribution histogram respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature hist (y), wherein y represents the grain distribution histogram.

Images