CN103632156B - Froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix - Google Patents

Abstract

The invention discloses a kind of froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix, first foam gray level image is carried out wavelet transformation, then respectively the wavelets approximation subgraph coefficient of different scale is carried out grey scale mapping, finally, obtain the multi-dimension texture feature of a kind of new reflection froth images grey scale change Frequency statistics rule according to neighboring gray level dependence matrix statistics, this feature has the robustness of higher reflection copper flotation production run state and is suitable to the separability of operating mode's switch。According to the textural characteristics obtained, the froth images of different operating modes can be made a distinction, reach effectively to identify the purpose of operating mode, and then control to provide Operating Guideline for flotation Optimizing manufacture。

Description

Froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix

Technical field

The invention belongs to image processing techniques and area of pattern recognition, particularly to a kind of froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix。

Background technology

Correct identification flotation operating mode is the basis and the key that realize the operation of flotation Optimizing manufacture。The visual signature of flotation froth contains a large amount of information relevant with production operation variable and product quality, is the important evidence judging flotation effect。In actual floatation process, mainly through the visual signature observing tank surface foam, operator judge that current working, this mode subjectivity and randomness are strong, have impact on the accurate judgement of operating mode。

Along with developing rapidly of the technology such as machine vision and image procossing, the work carrying out production status Intelligent Recognition in conjunction with flotation site froth images visual signature has made great progress。Research shows, the color of froth images, texture, size etc. are characterized by reflecting preferably flotation operating mode, and wherein, textural characteristics is due to insensitive and become application feature main in operating mode's switch to illumination variation。According to neighboring gray level dependence matrix (neighboringgrayleveldependencematrix, NGLDM) second degree statistics extracted has the spatial character such as clear and definite physical significance and tool rotational invariance, can reflect structure (thickness) characteristic of froth images texture preferably。But, utilize the feature that neighboring gray level dependence matrix extracts can only reflect the single scale space characteristics of froth images texture, it is impossible to catch the texture information of more horn of plenty from multiple yardsticks。

Therefore, it is necessary to design the extracting method of a kind of froth images multi-dimension texture feature。

Summary of the invention

The technical problem to be solved is to provide the extracting method of a kind of froth images multi-dimension texture feature, and the extracting method of this froth images multi-dimension texture feature has good pattern separability for froth images identification, and easy to implement。

A kind of froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix, comprises the following steps:

Step one: read RGB froth images according to the foam video that copper flotation site obtains, RGB froth images is carried out gray processing, then to foam gray level image I_p×q(x, y) carries out 2-d discrete wavelet decomposition, obtains the wavelet subband on each yardstick, and wavelet subband includes ll channel and details subgraph；

Wherein, p × q is the resolution of foam gray level image, and (x y) represents the coordinate of any pixel point in foam gray level image；

Step 2: the ll channel in the wavelet subband on each yardstick that step one is obtained carries out grey scale mapping, is mapped to the wavelet coefficient of ll channel the gray scale interval of 0 255 and is used for updating the wavelet coefficient of ll channel, obtain the ll channel after grey scale mappingWherein, when k × m is j for decomposition exponent number, the resolution of the ll channel obtained is decomposed；

Step 3: ask for the neighboring gray level dependence matrix of ll channel on each yardstick that step 2 obtains respectively；

Step 4: with the neighboring gray level dependence matrix of the ll channel on each yardstick that step 3 obtains, calculate fineness and 2 characteristic quantities of rugosity of every width ll channel, the fineness of the ll channel on all yardsticks and rugosity characteristic quantity are combined the textural characteristics as froth images。

Foam gray level image carrying out in described step one J rank 2-d discrete wavelet decompose, obtain the wavelet subband on each yardstick, J is integer, J >=2；

2-d discrete wavelet decomposition refers to image I to be decomposed^j(x, y) decomposes the details subgraph obtaining 1 ll channel and level, vertical and 3 different directions of diagonal, and 2-d discrete wavelet decomposition formula is as follows:

In formula,Represent the x continuous summation from 0 value to m-1 successively from 0 value to k-1 and y successively,For scaling function, Respectively 3 wavelet functions, whereinWithRepresenting the two scale equation of one-dimensional orthogonal multiresolution analysis and little wave equation respectively, when t is taken as α, i is taken as k；Accordingly, when t is taken as β, i is taken as m；AndRepresent that decomposing exponent number decomposes the detail coefficients of the Coefficients of Approximation of ll channel, the detail coefficients of horizontal direction details subgraph, the detail coefficients of vertical direction details subgraph and the diagonal details subgraph that obtain when being j respectively, the Mallat fast algorithm decomposed by 2-d wavelet obtains, and formula is as follows:

$c_{k,m}^j=\underset{l,n}{\Σ}{\stackrel{\‾}{h}}_{2k-l}{\stackrel{\‾}{h}}_{2m-n}{c}_{l,n}^{j+1}$

$d_{k,m}^{j,1}=\underset{l,n}{\Σ}{\stackrel{\‾}{h}}_{2k-1}-{\stackrel{\‾}{g}}_{2m-n}{c}_{l,n}^{j+1}$

$d_{k,m}^{j,2}=\underset{l,n}{\Σ}{\stackrel{\‾}{g}}_{2k-l}{\stackrel{\‾}{h}}_{2m-n}{c}_{l,n}^{j+1}$

$d_{k,m}^{j,3}=\underset{l,n}{\Σ}{\stackrel{\‾}{g}}_{2k-l}{\stackrel{\‾}{g}}_{2m-n}{c}_{l,n}^{j+1}$

In formula, k, m ∈ Z is system of representatives matrix number respectivelyRow and column, l, n ∈ Z respectively system of representatives matrix numberRow and column；Sequential for h is reversedCoefficient sequence h={h_iIt is a low pass filter, and h={-0.076 ,-0.030,0.498,0.804,0.298 ,-0.100 ,-0.013,0.032}；Sequential for g is reversedCoefficient sequence g={g_iIt is a high pass filter and g_i=(-1)ⁱh_1-i；

First rank are decomposed and are referred to when decomposing exponent number j=J, using foam gray level image as the object I being decomposed^j(x y) decomposes according to 2-d discrete wavelet decomposition formula；

The decomposition of z rank refers to when decomposing exponent number j=J-z+1, once decomposes the ll channel of acquisition in the past as the object I being decomposed^j(x, y) decomposes according to 2-d discrete wavelet decomposition formula, and z is integer, z >=2；

It is the coefficient matrix to the ll channel that froth images obtains at each Scale Decomposition。

Ll channel in the wavelet subband on each yardstick in described step 2, step one obtained carries out grey scale mapping and specifically refers to the coefficient matrix to each rank ll channelGrey scale mapping is carried out, it is thus achieved that the ll channel after grey scale mapping according to below equation

$M_{k\×m}^j(x,y)=255\×(c_{k,m}^j-c_{min})/(c_{max}-c_{min})$

Wherein, c_min、c_maxRepresent coefficient matrix respectivelyMinima in all elements and maximum。

Described step 3 is asked for the specifically comprising the following steps that of neighboring gray level dependence matrix of the coefficient matrix of ll channel

Step 1: obtain ll channelIn arbitrarily pixel (x, face neighborhood y)；

It is the digital picture of k × m to resolution(x, y) represents pixel coordinate in digital picture, x=0,1 ..., k-1, y=0,1 ..., m-1, (x, y) denotation coordination is (x, y) gray value of place's pixel, digital picture to fIn with pixel (x_c,y_c) centered by, D=1 is the face neighborhood definition of radius is:

$V_D^j(x_c,y_c)=\left\{(u,v)\right|(u,v)\&Element;M_{k\&times;m}^j,0<\mathrm{\&rho;}\left(\right(x_c,y_c),(u,v\left)\right)\&le;D\}$

Wherein, (u, v) | () } represent that (ρ represents two pixel (x for u, set v) formed for the element that meets under specified criteria ()_c,y_c) with (u, the distance between v):

ρ((x_c,y_c), (u, v))=max (| x_c-u|,|y_c-v|)

Step 2: build digital pictureThe gray scale frequency table of middle pixel

$P_D^j(x^{\&prime;},y^{\&prime;})=table\left(\right(g_c,s_c\left)\right)$

Wherein, table ((g_c,s_c)) represent a tables of data, (x ', y ') is the position coordinates of certain cell, x'=x in table_c-D, x'=0,1 ..., k-1-2D, y'=y_c-D, y'=0,1 ..., m-1-2D；(g_c,s_c) for the content of cell (x ', y '), g_c=f (x_c,y_c), s_cRepresent with pixel (x_c,y_c) centered by, D be radius face neighborhood in gray value be g_cPixel number, it may be assumed that

$s_c=\#\left\{(u,v)\right|(u,v)\&Element;V_D^j(x_c,y_c),f(u,v)=g_c\}$

# represent statistics set (u, v) | () } in meet element (u, number v) under specified criteria ()；

Step 3: calculate the neighboring gray level dependence matrix of each rank ll channel

Statistics (g_c,s_c) at gray scale frequency tableThe number of times of middle appearance, asks for each rank ll channel neighboring gray level dependence matrix respectivelyElementFor:

$q_D^j(g,s)=\#\left\{(g,s)\right|(g,s)\&Element;P_D^j(x^{\&prime;},y^{\&prime;})\},$

WhereinThe number of greyscale levels of ll channel is decomposed for corresponding jth rank, Represent that corresponding jth rank are decomposed in ll channel with pixel (x_c,y_c) centered by, radius be D face neighborhood in the number of pixel, namely

The fineness F of jth rank ll channel in described step 4_jComputing formula is as follows:

$F_j=\frac{\underset{g=0}{\overset{M_g^j-1}{\&Sigma;}}\underset{s=0}{\overset{N_D^j-1}{\&Sigma;}}\&lsqb;Q_D^j(g,s)/{(s+1)}^2\&rsqb;}{\underset{g=0}{\overset{M_g^j-1}{\&Sigma;}}\underset{s=0}{\overset{N_D^j-1}{\&Sigma;}}{Q}_D^j(g,s)}$

The rugosity C of jth rank ll channel in described step 4_jComputing formula is as follows:

$C_j=\frac{\underset{g=0}{\overset{M_g^j-1}{\&Sigma;}}\underset{s=0}{\overset{N_D^j-1}{\&Sigma;}}\&lsqb;{(s+1)}^2{Q}_D^j(g,s)\&rsqb;}{\underset{g=0}{\overset{M_g^j-1}{\&Sigma;}}\underset{s=0}{\overset{N_D^j-1}{\&Sigma;}}{Q}_D^j(g,s)}$

Wherein,The number of greyscale levels of ll channel is decomposed for corresponding jth rank, Represent that corresponding jth rank are decomposed in ll channel with pixel (x_c,y_c) centered by, radius be D face neighborhood in the number of pixel, namely Neighboring gray level dependence matrix for jth rank ll channel

The technology design of the present invention:

Wavelet transformation provides a kind of instrument researching and analysing texture on different scale, original image can be resolved into the sub-band images of different resolution, wherein low frequency subgraph can reflect the profile information of image preferably, it is thus achieved that the ratio statistical nature that single scale analysis is enriched more。But, utilize the statistical nature that wavelet multi-scale analysis is extracted to be difficult to describe structure (thickness) characteristic of froth images texture。

For this, the present invention is directed to neighboring gray level dependence matrix lack multiple dimensioned on texture information, and the problem that the statistical nature that small wave converting method extracts is difficult to describe structure (thickness) characteristic of froth images texture, propose a kind of new for multiple dimensioned neighboring gray level dependence matrix (multi-scaleneighboringgrayleveldependencematrix, M-NGLDM) froth images texture characteristic extracting method, for efficiently identifying copper flotation production status。

Beneficial effect

The invention provides a kind of froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix, froth images multi-dimension texture feature extracting method based on wavelet transformation, pass through Multiscale Image Processing Method, the characteristic vector obtained can well reflect the textural characteristics of each order of image, overcome neighboring gray level dependence matrix and the respective limitation of wavelet transformation in current froth images texture characteristic extracting method, namely neighboring gray level dependence matrix can only reflect the texture information on the single yardstick of froth images, lacks the description in foam texture multi-scale；The statistical nature avoiding multi-scale Wavelet Analysis extraction is difficult to describe the architectural characteristic of froth images texture。It is demonstrated experimentally that the texture characteristic amount that the present invention extracts has good pattern separability, it is possible to being separated in normal, aquation, three kinds of froth zones of viscosity well, and this method can directly realize on computers, cost is low, efficiency is high, it is easy to implement。

Accompanying drawing explanation

Fig. 1 is the froth images of three kinds of different operating modes, and wherein, figure a is normal froth images, and figure b is aquation froth images, and figure c is the image of viscous foam；

Fig. 2 is the dendrogram of normal foam gray level image 3 rank wavelet decomposition；

Fig. 3 is imageIn with pixel (x_c,y_c) centered by, D=1 be the face neighborhood schematic diagram of radius；

Fig. 4 is certain rank ll channelIntensity profile；

Fig. 5 is the gray scale frequency table corresponding to Fig. 4 intensity profile；

Fig. 6 is the neighboring gray level dependence matrix corresponding to Fig. 5 gray scale frequency table；

Fig. 7 is that feature distribution dissipates figure。

Detailed description of the invention

Below with reference to accompanying drawing be embodied as case the present invention is described in further details:

Below in conjunction with accompanying drawing, the specific embodiment of the present invention being described, certain copper flotation site has the foam of three kinds of different operating modes, respectively normal foam, aquation foam and viscous foam, and the froth images of these three difference operating mode is as shown in Figure 1。

The first step, reads RGB froth images according to the foam video that copper flotation site obtains, RGB froth images is carried out gray processing, then foam gray level image is carried out wavelet decomposition, obtain the wavelet subband on different scale；

Step 1: read RGB froth images by original foam video；

Step 2:RGB froth images gray processing。Foam gray level image I is become after original RGB froth images gray processing_600×800；

Step 3: select sym4 small echo to gray level image I_600×800Carrying out 3 rank decomposition, at each decomposition order, two-dimensional wavelet transformation will produce the details subgraph in 1 ll channel and level, vertical, 3 directions of diagonal。

In formula,Represent the x continuous summation from 0 value to m-1 successively from 0 value to k-1 and y successively,For scaling function, Respectively 3 wavelet functions, whereinWithRepresenting the two scale equation of one-dimensional orthogonal multiresolution analysis and little wave equation respectively, when t is taken as α, i is taken as k；Accordingly, when t is taken as β, i is taken as m；AndRepresent that decomposing exponent number decomposes the detail coefficients of the Coefficients of Approximation of ll channel, the detail coefficients of horizontal direction details subgraph, the detail coefficients of vertical direction details subgraph and the diagonal details subgraph that obtain when being j respectively, the Mallat fast algorithm decomposed by 2-d wavelet obtains, and formula is as follows:

In formula, k, m ∈ Z is system of representatives matrix number respectivelyRow and column, l, n ∈ Z respectively system of representatives matrix numberRow and column；Sequential for h is reversedCoefficient sequence h={h_iIt is a low pass filter, and h={-0.076 ,-0.030,0.498,0.804,0.298 ,-0.100 ,-0.013,0.032}；Sequential for g is reversedCoefficient sequence g={g_iIt is a high pass filter and g_i=(-1)ⁱh_1-i；

Foam gray level image carries out in this example three rank decomposition, and namely the value of J=3, j is followed successively by 3, and 2,1, first application formula 1 carries out the two-dimensional wavelet transformation on the first rank, and (taking j=3, k=300, m=400, l=600, n=800) obtains 1 ll channelWith 3 details subgraphsIts coefficient is calculated by formula 2, nowIt is the 2-D gray image matrix I that step 1 obtains_p×q=I_600×800(Calculation procedure be: to arbitrary fixing columns n=800, first useWithEach column vector make convolution, carry out downward two sampling and obtainThen use againWithEach row vector make convolution, carry out downward two sampling and obtain)；Then, recycling formula 1 is to single order ll channel (low frequency part)Proceed the two-dimensional wavelet transformation (now j=2, k=150, m=200, l=300, n=400) of second-order, again obtain 1 ll channelWith 3 details subgraphsThe rest may be inferred, namely obtains3 decompose orders wavelet transformations, correspondingly, just obtain gray level image I_600×800Multi-scale Representation；Fig. 2 is the dendrogram of normal froth images 3 rank wavelet decomposition；

Second step, adopts formula 3 to each rank ll channel coefficientCarry out grey scale mapping so that it is be mapped in the gray scale interval of 0 255, obtain new each level matrix number

Wherein, c_min、c_maxRepresent coefficient matrix respectivelyMinima in all elements and maximum。

3rd step, to each rank ll channel wavelet coefficient after grey scale mappingAsk for neighboring gray level dependence matrix respectively。

Step 1: adopt formula 4 and formula 5 to define each rank digital pictureIn the field, face of certain central pixel point。

It is the digital picture of k × m to resolution(x, y) represents pixel coordinate in digital picture, x=0,1 ..., k-1, y=0,1 ..., m-1, (x, y) denotation coordination is (x, y) gray value of place's pixel, digital picture to fIn with pixel (x_c,y_c) centered by, D=1 is the face neighborhood definition of radius is:

Wherein, (u, v) | () } represent that (ρ represents two pixel (x for u, set v) formed for the element that meets under specified criteria ()_c,y_c) with (u, the distance between v):

ρ((x_c,y_c), (u, v))=max (| x_c-u|,|y_c-v |) formula 5

As D=1, imageIn certain central pixel point (x_c,y_c) face neighborhood be remove central pixel point 3 × 3 regions, i.e. the set of all pixels on 8-neighborhood direction, as shown in Figure 3。

Step 2: adopt formula 6 and formula 7 to ask for each rank digital picture Gray scale frequency tableP₁ ³, P₁ ², P₁ ¹；

Gray scale frequency table

Wherein table ((g_c,s_c)) represent a tables of data, (x ', y ') is the position coordinates of certain cell, x'=x in table_c-D, x'=0,1 ..., k-1-2D, y'=y_c-D, y'=0,1 ..., m-1-2D；(g_c,s_c) for the content of cell (x ', y '), g_c=f (x_c,y_c), s_cRepresent pixel (x_c,y_c) the face neighborhood that radius is D in gray value be g_cPixel number, it may be assumed that

# represent statistics set (u, v) | () } in meet element (u, number v) under specified criteria ()。

P₁ ³Middle x_c=0,1 ..., 300-1, y_c=0,1 ..., 400-1, x'=x_c-1, x'=0,1 ..., 300-1-2, y'=y_c-1, y'=0,1 ..., 400-1-2；P₁ ²Middle x_c=0,1 ..., 150-1, y_c=0,1 ..., 200-1, x'=x_c-1, x'=0,1 ..., 150-1-2, y'=y_c-1, y'=0,1 ..., 200-1-2；P₁ ¹Middle x_c=0,1 ..., 75-1, y_c=0,1 ..., 100-1, x '=x_c-1, x '=0,1 ..., 75-1-2, y '=y_c-1, y '=0,1 ..., 100-1-2。

Step 3: adopt formula 8 to add up (g_c,s_c) at gray scale frequency tableThe number of times of middle appearance, asks for each rank ll channel neighboring gray level dependence matrix respectivelyElement

WhereinThe number of greyscale levels of ll channel is decomposed on each rankAs radius D=1, certain central pixel point (x in ll channel is decomposed on each rank_c,y_c) pixel number in the neighborhood of face is

Fig. 4 gives certain rank ll channelIntensity profile example, formula 6 and formula 7 calculate 1 (k-1-2) × (m-1-2) the gray scale frequency table P of this rank subgraph when obtaining D=1₁ ^j(x ', y '), as shown in Figure 5；Multiple dimensioned neighboring gray level dependence matrix is obtained again by formula 8 statisticsAs shown in Figure 6。

4th step, obtains neighboring gray level dependence matrix striked by the ll channel of each rankFormula 9 and formula 10 is adopted to calculate fineness F_jWith rugosity C_j2 characteristic quantities, respectively as the multi-dimension texture feature of froth images after normalization。

J rank ll channel to a width close grain image, in the row that in the element set that in multiple dimensioned neighboring gray level dependence matrix, numerical value is bigger, s value is less in a matrix, namely in left-hand column, this makes less s'sIt is worth bigger。Therefore, the F of piece image_jBeing worth more big, the texture of image is more thin。

To each rank ll channel neighboring gray level dependence matrixCalculate its rugosity C_jValue (Coarseness), computing formula is:

J rank ll channel to a thicker image of width texture, in the row that in the element set that in multiple dimensioned neighboring gray level dependence matrix, numerical value is bigger, s value is bigger in a matrix, namely in right-hand column, this makes bigger s'sIt is worth also big。Therefore, the C of piece image_jBeing worth more big, the texture of image is more thick。

The froth images texture feature vector scatter diagram finally given is showed, as shown in Figure 7, giving the first rank (j=3) and second-order (j=2) froth images textural characteristics scatter diagram during j=3, wherein second-order fineness feature is by the cartographic represenation of area of round dot out。Investigating the degree that the froth zone of different operating modes is separated by extracted characteristic quantity, as seen from Figure 7, the texture characteristic amount that the present invention extracts has good pattern separability, it is possible to separated in normal, aquation, three kinds of froth zones of viscosity well。

Claims (5)

1. the froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix, it is characterised in that comprise the following steps:

Step one: read RGB froth images according to the foam video that copper flotation site obtains, RGB froth images is carried out gray processing, then to foam gray level image I_p×q(x, y) carries out 2-d discrete wavelet decomposition, obtains the wavelet subband on each yardstick, and wavelet subband includes ll channel and details subgraph；

Wherein, p × q is the resolution of foam gray level image, and (x y) represents the coordinate of any pixel point in foam gray level image；

Step 2: the ll channel in the wavelet subband on each yardstick that step one is obtained carries out grey scale mapping, is mapped to the wavelet coefficient of ll channel the gray scale interval of 0 255 and is used for updating the wavelet coefficient of ll channel, obtain the ll channel after grey scale mappingWherein, when k × m is j for decomposition exponent number, the resolution of the ll channel obtained is decomposed；

Step 3: ask for the neighboring gray level dependence matrix of ll channel on each yardstick that step 2 obtains respectively；

Step 4: with the neighboring gray level dependence matrix of the ll channel on each yardstick that step 3 obtains, calculate fineness and 2 characteristic quantities of rugosity of every width ll channel, the fineness of the ll channel on all yardsticks and rugosity characteristic quantity are combined the textural characteristics as froth images。

2. the froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix according to claim 1, it is characterized in that, foam gray level image is carried out J rank 2-d discrete wavelet by described step one and decomposes, obtain the wavelet subband on each yardstick, J is integer, J >=2；

2-d discrete wavelet decomposition refers to image I to be decomposed^j(x, y) decomposes the details subgraph obtaining 1 ll channel and level, vertical and 3 different directions of diagonal, and 2-d discrete wavelet decomposition formula is as follows:

In formula,Represent the x continuous summation from 0 value to m-1 successively from 0 value to k-1 and y successively,For scaling function, Respectively 3 wavelet functions, whereinWithRepresent the two scale equation of one-dimensional orthogonal multiresolution analysis and little wave equation respectively；AndRepresent that decomposing exponent number decomposes the detail coefficients of the Coefficients of Approximation of ll channel, the detail coefficients of horizontal direction details subgraph, the detail coefficients of vertical direction details subgraph and the diagonal details subgraph that obtain when being j respectively, the Mallat fast algorithm decomposed by 2-d wavelet obtains, and formula is as follows:

In formula, k, m ∈ Z is system of representatives matrix number respectivelyRow and column, l, n ∈ Z respectively system of representatives matrix numberRow and column；Sequential for h is reversedCoefficient sequence h={h_iIt is a low pass filter, and h={-0.076 ,-0.030,0.498,0.804,0.298 ,-0.100 ,-0.013,0.032}；Sequential for g is reversedCoefficient sequence g={g_iIt is a high pass filter and g_i=(-1)ⁱh_1-i；

First rank are decomposed and are referred to when decomposing exponent number j=J, using foam gray level image as the object I being decomposed^j(x y) decomposes according to 2-d discrete wavelet decomposition formula；

The decomposition of z rank refers to when decomposing exponent number j=J-z+1, once decomposes the ll channel of acquisition in the past as the object I being decomposed^j(x, y) decomposes according to 2-d discrete wavelet decomposition formula, and z is integer, z >=2；

It is the coefficient matrix to the ll channel that froth images obtains at each Scale Decomposition。

3. the froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix according to claim 2, it is characterized in that, the ll channel in the wavelet subband on each yardstick in described step 2, step one obtained carries out grey scale mapping and specifically refers to the coefficient matrix to each rank ll channelGrey scale mapping is carried out, it is thus achieved that the ll channel after grey scale mapping according to below equation

Wherein, c_min、c_maxRepresent coefficient matrix respectivelyMinima in all elements and maximum。

4. the froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix according to claim 3, it is characterised in that ask for the specifically comprising the following steps that of neighboring gray level dependence matrix of the coefficient matrix of ll channel in described step 3

Step 1: obtain ll channelIn arbitrarily pixel (x, face neighborhood y)；

It is the digital picture of k × m to resolution(x, y) represents pixel coordinate in digital picture, x=0,1 ..., k-1, y=0,1 ..., m-1, (x, y) denotation coordination is (x, y) gray value of place's pixel, digital picture to fIn with pixel (x_c,y_c) centered by, D=1 is the face neighborhood definition of radius is:

Wherein, (u, v) | () } represent meet element under specified criteria () (u, set v) formed, ρ represent two pixels (xc, yc) with (u, the distance between v):

ρ((x_c,y_c), (u, v))=max (| x_c-u|,|y_c-v|)

Step 2: build digital pictureThe gray scale frequency table of middle pixel

Wherein, table ((g_c,s_c)) represent a tables of data, (x ', y ') is the position coordinates of certain cell, x'=x in table_c-D, x'=0,1 ..., k-1-2D, y'=y_c-D, y'=0,1 ..., m-1-2D；(g_c,s_c) for the content of cell (x ', y '), g_c=f (x_c,y_c), s_cRepresent with pixel (x_c,y_c) centered by, D be radius face neighborhood in gray value be g_cPixel number, it may be assumed that

# represent statistics set (u, v) | () } in meet element (u, number v) under specified criteria ()；

Step 3: calculate the neighboring gray level dependence matrix of each rank ll channel

Statistics (g_c,s_c) at gray scale frequency tableThe number of times of middle appearance, asks for each rank ll channel neighboring gray level dependence matrix respectivelyElementFor:

Wherein The number of greyscale levels of ll channel is decomposed for corresponding jth rank, Represent that corresponding jth rank are decomposed in ll channel with pixel (x_c,y_c) centered by, radius be D face neighborhood in the number of pixel, namely

5. the froth images texture characteristic extracting method based on multiple dimensioned neighborhood correlation matrix according to any one of claim 1-4, it is characterised in that the fineness F of jth rank ll channel in described step 4_jComputing formula is as follows:

The rugosity C of jth rank ll channel in described step 4_jComputing formula is as follows:

Wherein, The number of greyscale levels of ll channel is decomposed for corresponding jth rank, Represent that corresponding jth rank are decomposed in ll channel with pixel (x_c,y_c) centered by, radius be D face neighborhood in the number of pixel, namely Neighboring gray level dependence matrix for jth rank ll channel