CN105405149A - Composite texture feature extraction method for flotation froth image - Google Patents

Abstract

The present invention discloses a composite texture feature extraction method for a flotation froth image. The method comprises: firstly, in a grayscale quantization matrix of a froth image, acquiring a face neighborhood set of all central pixel points; then, for all the central pixel points, constructing a three-dimensional data table and obtaining a nested grayscale frequency table; again, acquiring an improved neighborhood grayscale correlation matrix; and finally, obtaining a new composite texture feature, wherein the feature integrates a size, a texture and a roughness degree of froth, and has relatively high stability and separability in reflecting a texture of flotation froth; and according to the extracted composite texture feature, it is easy to distinguish flotation froth images with different operating conditions in different ore grades, thereby having a relatively high accuracy rate of recognizing operating conditions. The composite texture feature extraction method for the flotation froth image provided by the present invention is simple and effective, and is very important to guide the recognition of froth operating conditions in a mineral flotation site.

Description

A kind of compounded texture feature extracting method of floatation foam image

Technical field

The invention belongs to image procossing, pattern-recognition and mineral floating technical field, particularly a kind of compounded texture feature extracting method of floatation foam image.

Background technology

Froth appearance feature is the concentrated expression of mineral floating operating mode, is considered to closely related with the quality of flotation effect.How accurately to extract froth appearance feature closely-related with crucial production target in floatation process, be the key realizing flotation operating mode's switch.For a long time, on-the-spot experienced workman, by observing the status adjustment operating mode of foam surface, because artificial randomness and subjectivity cause flotation state labile, is difficult to be adjusted to optimum state, causes mineral wealth utilization factor not high, cause the wasting of resources.

Floatation process in recent years based on machine vision controls and optimizes to become domestic and international study hotspot, and the operating mode's switch wherein based on machine vision is one of main contents of research.Froth appearance feature and flotation operating mode closely related, be the indicator of flotation performance.The froth appearance feature being usually used in flotation intelligence operating mode's switch mainly contains foam color, size, texture etc., wherein, textural characteristics is as one of most important feature of froth appearance, combine the features such as the size and shape of foam, have not with features such as ambient light impacts simultaneously, be widely used in flotation operating mode's switch.Neighboring gray level dependence matrix is a kind of texture characteristic extracting method of conventional Corpus--based Method, and it has rotational invariance, not with illumination variation impact and the feature such as computing velocity is fast.The secondary statistical nature such as rugosity and fineness based on neighboring gray level dependence matrix can reflect the texture feature of foam surface preferably.Because traditional neighboring gray level dependence matrix method only considers the central pixel point number identical or close with its neighborhood territory pixel point gray-scale value, but the number that differs between them and phase extent is not considered, therefore, lost the space distribution attribute that can reflect image pixel diversity factor in a large number, be difficult to the texture information of accurately reflection foam surface comprehensively.In addition, different enter ore deposit grade condition under, froth images also has larger difference, and traditional neighboring gray level dependence matrix and do not consider the texture characteristic extracting method into ore deposit product position influence, is difficult to the demand adapting to accurately identify froth flotation operating mode.

Therefore, be necessary the compounded texture feature extracting method designing a kind of floatation foam image, so as more fully to catch different enter foam surface texture information under the grade condition of ore deposit.

Summary of the invention

Technical matters to be solved by this invention is to provide a kind of compounded texture feature extracting method of floatation foam image, the method can reflect that difference enters the difference of the froth images textural characteristics under the grade of ore deposit more all sidedly, there is good characteristic validity and separability, to be suitable for the needs of follow-up operating mode's switch accuracy.

A compounded texture feature extracting method for floatation foam image, comprises the following steps:

Step one: read RGB froth images according to the foam video that froth flotation scene obtains, RGB image is carried out gray processing, obtains gray level image matrix; Gray level image is quantized, obtains quantization matrix; Obtain the face Neighbourhood set of all central pixel point in quantization matrix;

Step 2: the absolute difference calculating pixel and central pixel point gray-scale value in each neighborhood; Add up the number of each absolute difference; For central pixel point all in froth images, build three-dimensional data table, in described three-dimensional data table, element is respectively central pixel point gray-scale value, each absolute difference number, absolute difference;

Step 3: build the nested gray scale frequency table based on three-dimensional data table;

Step 4: according to constructed nested gray scale frequency table, obtain a kind of neighboring gray level dependence matrix of improvement;

Step 5: according to obtained improvement neighboring gray level dependence matrix, asks for a kind of new secondary statistical nature---compounded texture.

The face Neighbourhood set concrete steps of all central pixel point of froth images matrix after the acquisition in described step one quantizes are as follows:

The foam video that step 1 obtains according to froth flotation scene reads RGB froth images, RGB image is carried out gray processing, obtains gray level image matrix A (x, y):

A _k×m(x,y)＝0.290×R(x,y)+0.587×G(x,y)+0.114×B(x,y)

In formula, k × m is the resolution of foam gray level image, and (x, y) represents the coordinate of arbitrary pixel in foam gray level image, x=0,1 ..., k-1, y=0,1 ..., m-1; R (x, y), G (x, y), B (x, y) are respectively R, G, B matrix of froth images.

Step 2 pair gray level image matrix A (x, y) quantizes, and obtains quantization matrix M (x, y):

M in formula _gfor the quantification progression of froth images, for rounding downwards.

Step 3 in quantization matrix M (x, y), with all central pixel point (x _c, y _c) centered by, D is radius, obtains face Neighbourhood set:

V _D(x _c,y _c)＝{(u,v)|(u,v)∈M _k×m,0＜ρ((x _c,y _c),(u,v))≤D}

Wherein x _c∈ D, D+1 ..., k-1-D, y _c∈ D, D+1 ..., m-1-D; { (u, v) | () } represents the set that the point (u, v) met under specified criteria () forms, ρ ((x _c, y _c), (u, v)) pixel (u, v) and central pixel point (x in presentation surface neighborhood _c, y _c) between distance:

ρ((x _c,y _c),(u,v))＝max(|x _c-u|,|y _c-v|)

Max in formula (| x _c-u|, | y _c-v|) represent pixel distance | x _c-u|, | y _cmaximal value in-v|.

The three-dimensional data table concrete steps building froth images central pixel point in described step 2 are as follows:

Step 1 calculates face Neighbourhood set V _d(x _c, y _c) in pixel (u, v) gray-scale value and central pixel point (x in each neighborhood _c, y _c) the absolute difference i of gray-scale value:

i＝|f(u,v)-f(x _c,y _c)|＝|f(u,v)-g _c|

F (u, v) and f (x in formula _c, y _c)=g _crepresent pixel (u, v) and central pixel point (x respectively _c, y _c) gray-scale value, g _c=0,1 ..., M _g-1, i=0,1 ..., M _g-1.

Step 2: statistics absolute difference is the number of i

$s_{g_c,i}=\#\left\{(u,v)\right|(u,v)\∈V_D(x_c,y_c),i\}$

In formula, # represents in statistics set { (u, v) | () } number of the point (u, v) met under specified criteria ().

Step 3: the three-dimensional data table building each central pixel point of froth images;

Define certain central pixel point (x _c, y _c) three-dimensional data table f _d(x _c, y _c, i) as follows:

$f_D(x_c,y_c,i)=table(g_c,s_{g_c,i},i)$

In formula be that in a table, element is respectively g _c, i, size is 1 × 1 × M _gthree-dimensional data table.

For central pixel point all in froth images, build corresponding three-dimensional data table.

The nested gray scale frequency table method built in described step 3 based on three-dimensional data table is as follows:

Define nested gray scale frequency table F _d(x ' _c, y ' _c, i) as follows:

F _D(x _c',y _c',i)＝table(f _D(x _c,y _c,i))

X in formula _c'=x _c-D, x _c' ∈ 0,1 ..., k-1-2D, y _c'=y _c-D, y _c' ∈ 0,1 ..., m-1-2D.Nested gray scale frequency table F _d(x ' _c, y ' _c, certain cell location (x i) _c', y _c') element be

By step 2 for all central pixel point (x _c, y _c) constructed by each three-dimensional data table f _d(x _c, y _c, i), embed respective cells position in nested gray scale frequency table, construct a size for (k-2D) × (m-2D) × M _gnested gray scale frequency table.

A kind of neighboring gray level dependence matrix of improvement is obtained in described step 4:

Definition improves neighborhood correlation matrix Q _d(g, i) is as follows:

G=0 in formula, 1 ..., M _g-1, i=0,1 ..., M _g-1, q _d(g, i) is matrix Q _d(g, i) is at the element at position (g, i) place.

At the nested gray scale frequency table F that step 3 obtains _d(x ' _c, y ' _c, i), pixel gray-scale value and central pixel point gray-scale value g in the neighborhood of statistics face _cbetween absolute difference all when being i summation q _d(g _c, i):

In formula, Σ represents set { s _g,i| () } in meet all s of specified criteria () _g,Isum, & & presentation logic and computing.

By q _d(g _c, i) as matrix Q _d(g, i) is at position (g=g _c, the i) element at place, obtains a M _g× M _gimprovement neighboring gray level dependence matrix Q _d(g, i).

A kind of new compounded texture feature is calculated in described step 5:

According to the neighboring gray level dependence matrix improved, define a kind of new secondary statistical nature, namely compounded texture feature CT is:

$CT=\frac{\underset{g=0}{\overset{M_g-1}{\Σ}}{g}^2\left(\underset{i=0}{\overset{M_g-1}{\Σ}}{Q}_D\right(g,i\left)\right)}{\underset{g=0}{\overset{M_g-1}{\Σ}}\underset{i=0}{\overset{M_g-1}{\Σ}}\[{(i+1)}^2{Q}_D(g,i)\]}.$

Beneficial effect

The invention provides a kind of compounded texture feature extracting method of floatation foam image, it is relative to traditional neighboring gray level dependence matrix, not only add up the number that neighborhood territory pixel point is identical with central pixel point, and added up different size and corresponding number between the two, more fully can reflect the texture information of foam surface.Meanwhile, on the basis of the neighboring gray level dependence matrix improved, defining a kind of new secondary statistical nature---compounded texture feature, it combines the size of foam, texture and roughness, can describe the architectural characteristic of froth images textural characteristics more comprehensively.Experiment proves, the new compounded texture feature that the present invention extracts, has better characteristic validity, separability, and can reflect that difference enters the variation relation between concentrate grade under the grade condition of ore deposit and foam textural characteristics.Because computing velocity is fast, the ONLINE RECOGNITION of mineral floating operating mode can be effective to.

Accompanying drawing explanation

Fig. 1 is the froth images of 5 kinds of different operating modes, and wherein, figure a is betterness, and figure b is the grade such as good, and figure c is medium grade, and figure d is deviation grade, and figure e is extremely low-grade;

Pixel (x centered by Fig. 2 _c, y _c) face neighborhood schematic diagram (D=1);

Fig. 3 is simple description figure (D=1, the M that improve neighboring gray level dependence matrix solution procedure _g=8); Wherein, the grey level quantization matrix A that (a) is image is schemed _{k × m}(x, y); Figure (b) is three-dimensional data table f _d(x _c, y _c, i); Figure (c) is nested gray scale frequency table F _d(x _c', y _c', i); Figure (d) is the neighboring gray level dependence matrix Q improved _d(g, i);

Fig. 4 is the compounded texture characteristic interval distribution plan under each concentrate grade grade;

Fig. 5 is under entering ore deposit grade grade and being 2 conditions, and compounded texture feature is for the distribution plan of 5 kinds of different operating modes.

Embodiment

Below with reference to accompanying drawing and concrete case study on implementation, the present invention is described in further details.

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is described, there is the foam of 5 kinds of different nominal situations at certain antimony scene of roughly selecting, they divide according to concentrate grade grade, these 5 kinds of grades are respectively: excellent (product place value >=40.5), good grade (product place value: [37.5-40.5)), medium (product place value: [33-37.5)), deviation (product place value: [and 27.5-33)), extreme difference (product place value <27.5).The froth images of 5 kinds of different operating modes as shown in Figure 1.Enter ore deposit grade simultaneously and can be divided into 5 etc. according to the experience that on-the-spot antimony roughly selects workman, they are respectively: high (product place value >=2.14), higher (product place value: [1.75.-2.14)), medium (product place value: [1.33-1.75)), lower (product place value: [1.08-1.33)), low grade (product place value <1.08), as shown in the numeral in Fig. 1 froth images lower left corner, 1,2 ... 5 represent respectively into ore deposit grade to be high, it is higher ..., low etc.Can be found by froth images, under difference enters ore deposit grade condition, even if corresponding identical concentrate grade, the froth images that antimony is roughly selected still can present notable difference.

The first step, roughly select foam video according to the antimony that golden antimony flotation site collects, read RGB froth images, RGB froth images is converted into and carries out gray matrix, carrying out 16 grades of quantifications to gray matrix, is namely that the matrix quantization of 0 ~ 255 is the matrix of 0 ~ 15 to gray-scale value by gray-scale value.Finally obtain the face Neighbourhood set of all central pixel point of the froth images matrix after quantizing, concrete steps are as follows:

The foam video that step 1 obtains according to froth flotation scene reads RGB froth images, RGB image is carried out gray processing, obtains gray level image matrix A (x, y):

A _{k × m}(x, y)=0.290 × R (x, y)+0.587 × G (x, y)+0.114 × B (x, y) formula 1

In formula, k × m (k=600, m=800) is the resolution of foam gray level image, and (x, y) represents the coordinate of arbitrary pixel in foam gray level image, x=0,1 ..., k-1, y=0,1 ..., m-1; R (x, y), G (x, y), B (x, y) are respectively R, G, B matrix of froth images.

Step 2 pair gray level image matrix A (x, y) carries out 16 grades of quantifications, obtains quantization matrix M (x, y):

formula 2

M in formula _g(M _g=16) be the quantification progression of froth images, for rounding downwards.

Step 3 in quantization matrix M (x, y), with all central pixel point (x _c, y _c) centered by, the face Neighbourhood set of radius D=1 is:

V _d(x _c, y _c)={ (u, v) | (u, v) ∈ M _{k × m}, 0 < ρ ((x _c, y _c), (u, v))≤D} formula 3

Wherein x _c∈ D, D+1 ..., k-1-D, y _c∈ D, D+1 ..., m-1-D; { (u, v) | () } represents the set that the point (u, v) met under specified criteria () forms, ρ ((x _c, y _c), (u, v)) pixel (u, v) and central pixel point (x in presentation surface field _c, y _c) between distance:

ρ ((x _c, y _c), (u, v))=max (| x _c-u|, | y _c-v|) formula 4

Max in formula (| x _c-u|, | y _c-v|) represent pixel distance | x _c-u|, | y _cmaximal value (as shown in Figure 2) in-v|.

Second step, build the three-dimensional data table that antimony roughly selects picture centre pixel, concrete steps are as follows:

Step 1 calculates face Neighbourhood set V _d(x _c, y _c) in pixel (u, v) gray-scale value and central pixel point (x in each neighborhood _c, y _c) the absolute difference i of gray-scale value:

I=|f (u, v)-f (x _c, y _c) |=| f (u, v)-g _c| formula 5

F (u, v) and f (x in formula _c, y _c)=g _crepresent pixel (u, v) and central pixel point (x respectively _c, y _c) gray-scale value, g _c=0,1 ..., M _g-1, i=0,1 ..., M _g-1.

Step 2 adds up the number that absolute difference is i

$s_{g_c,i}=\#\left\{(u,v)\right|(u,v)\&Element;V_D(x_c,y_c),i\}$ Formula 6

In formula, # represents in statistics set { (u, v) | () } number of the point (u, v) met under specified criteria ().

Step 3: the three-dimensional data table building each central pixel point of froth images;

Define certain central pixel point (x _c, y _c) three-dimensional data table f _d(x _c, y _c, i) as follows:

$f_D(x_c,y_c,i)=table(g_c,s_{g_c,i},i)$ Formula 7

In formula be that in a table, element is respectively g _c, i, size is the three-dimensional data table of 1 × 1 × 15, and i gets 0,1 successively ..., 15.

For central pixel point all in froth images, build corresponding three-dimensional data table.The face Neighbourhood set of central pixel point is as shown in Fig. 3 (a), and corresponding three-dimensional tables of data is as shown in Fig. 3 (b).

3rd step, builds the nested gray scale frequency table that antimony roughly selects image;

Define nested gray scale frequency table F _d(x ' _c, y ' _c, i) as follows:

F _d(x _c', y _c', i)=table (f _d(x _c, y _c, i)) and formula 8

X in formula _c'=x _c-D, x _c' ∈ 0,1 ..., k-1-2D, y _c'=y _c-D, y _c' ∈ 0,1 ..., m-1-2D.Nested gray scale frequency table F _d(x ' _c, y ' _c, certain cell location (x i) _c', y _c') element be

By second step for all central pixel point (x _c, y _c) constructed by each three-dimensional data table f _d(x _c, y _c, i), embed respective cells position in nested gray scale frequency table, construct the nested gray scale frequency table that a size is 588 × 788 × 16.As shown in Fig. 3 (c).

4th step, obtains a kind of neighboring gray level dependence matrix of improvement:

Definition improves neighborhood correlation matrix Q _d(g, i) is as follows:

formula 9

G=0 in formula, 1 ..., M _g-1, i=0,1 ..., M _g-1, q _d(g, i) is matrix Q _d(g, i) is at the element at position (g, i) place.

At the nested gray scale frequency table F that the 3rd step obtains _d(x ' _c, y ' _c, i), pixel gray-scale value and central pixel point gray-scale value g in the neighborhood of statistics face _cbetween absolute difference all when being i summation q _d(g _c, i):

formula 10

In formula, Σ represents set { s _g,i| () } in meet all s of specified criteria () _g,Isum, & & presentation logic and computing.

By q _d(g _c, i) as matrix Q _d(g, i) is at position (g=g _c, the i) element at place, obtains the improvement neighboring gray level dependence matrix Q of 16 × 16 _d(g, i).As shown in Fig. 3 (d).

5th step, calculates a kind of new compounded texture feature:

According to the neighboring gray level dependence matrix improved, define a kind of new secondary statistical nature, namely compounded texture feature CT is:

$CT=\frac{\underset{g=0}{\overset{M_g-1}{\&Sigma;}}{g}^2\left(\underset{i=0}{\overset{M_g-1}{\&Sigma;}}{Q}_D\right(g,i\left)\right)}{\underset{g=0}{\overset{M_g-1}{\&Sigma;}}\underset{i=0}{\overset{M_g-1}{\&Sigma;}}\&lsqb;{(i+1)}^2{Q}_D(g,i)\&rsqb;}$ Formula 11

The molecule of compounded texture feature reflects and average characteristics, and when the higher i.e. antimony content of concentrate grade is higher, molecule is larger; The fineness degree of denominator reflection image texture, concentrate grade is higher, and when the texture of image is thinner, compounded texture eigenwert is larger.CT is compounded with fineness degree feature and average characteristics, can reflect the information such as the size on foam top layer, texture and roughness well.

Fig. 4 shows the compounded texture characteristic interval distribution under each concentrate grade grade, as can be seen here, when being in different concentrate grade grades, compounded texture eigenwert is distributed in different intervals, and when concentrate grade grade is higher, compounded texture eigenwert is maximum, therefore, compounded texture feature can reflect the change of concentrate grade, can as the indicator of concentrate grade.

Figure 5 provides entering ore deposit grade grade is 2 (higher), and the compounded texture feature extracted is for the distribution plan of 5 kinds of different operating modes.As seen from Figure 5, identical enter under the grade condition of ore deposit, different operating modes can obviously make a distinction by compounded texture feature.Therefore, compounded texture feature has good pattern separability, may be used for operating mode's switch, for antimony rougher process controls to provide guidance with optimization.

Neighboring gray level dependence matrix has Space Rotating unchangeability, not with illumination effect and the advantage such as computing velocity is fast, just in time agree with flotation site environment more complicated and the huge feature of flotation data volume, make it to become a kind of conventional texture characteristic extracting method in froth flotation.But traditional neighboring gray level dependence matrix only considered the neighborhood territory pixel point number identical or close with central pixel point, but the number that differs between them and phase extent is not considered, therefore, lost the space distribution attribute that can reflect image pixel diversity factor in a large number, be difficult to the texture information comprehensively reflecting foam surface.

Therefore, the present invention is directed to traditional neighboring gray level dependence matrix and have ignored the problem that there are differences between neighborhood territory pixel point and central pixel point, a kind of compounded texture feature extracting method of floatation foam image is proposed, the method has not only added up the neighborhood territory pixel point number identical with central pixel point, and has added up different size and corresponding number between the two.On the basis of the neighboring gray level dependence matrix improved, a kind of new textural characteristics---compounded texture feature of the size, texture and the roughness that combine foam is proposed, enter foam surface texture information under the grade condition of ore deposit more fully to express difference, be more easy to the accurate identification of froth flotation operating mode.

Claims (6)

1. a compounded texture feature extracting method for floatation foam image, is characterized in that, comprise the following steps:

Step one: read RGB froth images according to the foam video that mineral froth flotation site obtains, RGB froth images is carried out gray processing, obtains gray level image matrix; Gray level image is quantized, obtains quantization matrix; Obtain the face Neighbourhood set of all central pixel point in quantization matrix;

Step 2: the absolute difference calculating pixel and central pixel point gray-scale value in each neighborhood; Add up the number of each absolute difference; For central pixel point all in image, build three-dimensional data table, in described three-dimensional data table, element is respectively central pixel point gray-scale value, each absolute difference number, absolute difference;

Step 3: build the nested gray scale frequency table based on three-dimensional data table;

Step 4: according to constructed nested gray scale frequency table, obtain a kind of neighboring gray level dependence matrix of improvement;

Step 5: according to obtained improvement neighboring gray level dependence matrix, asks for a kind of new compounded texture feature.

2. the compounded texture feature extracting method of floatation foam image according to claim 1, is characterized in that, described step one concrete steps are as follows:

Step 1: read RGB froth images according to the foam video that froth flotation scene obtains, RGB image is carried out gray processing, obtains gray level image matrix A (x, y):

A _k×m(x,y)＝0.290×R(x,y)+0.587×G(x,y)+0.114×B(x,y)

In formula, k × m is the resolution of foam gray level image, and (x, y) represents the coordinate of arbitrary pixel in foam gray level image, x=0,1 ..., k-1, y=0,1 ..., m-1; R (x, y), G (x, y), B (x, y) are respectively R, G, B matrix of froth images;

Step 2: quantize gray level image matrix A (x, y), obtains quantization matrix M (x, y):

M in formula _gfor the quantification progression of froth images, for rounding downwards;

Step 3: in quantization matrix M (x, y), with all central pixel point (x _c, y _c) centered by, D is radius, obtains face Neighbourhood set:

V _D(x _c,y _c)＝{(u,v)|(u,v)∈M _k×m,0＜ρ((x _c,y _c),(u,v))≤D}

Wherein, x _c∈ D, D+1 ..., k-1-D; y _c∈ D, D+1 ..., m-1-D, { (u, v) | () } represents the set that the point (u, v) met under specified criteria () forms, ρ ((x _c, y _c), (u, v)) represent pixel (u, v) and central pixel point (x in neighborhood _c, y _c) between distance:

ρ((x _c,y _c),(u,v))＝max(|x _c-u|,|y _c-v|)

Max in formula (| x _c-u|, | y _c-v|) represent pixel distance | x _c-u|, | y _cmaximal value in-v|.

3. the compounded texture feature extracting method of floatation foam image according to claim 2, is characterized in that, described step 2 concrete steps are as follows:

Step 1: calculate face Neighbourhood set V _d(x _c, y _c) in pixel (u, v) gray-scale value and central pixel point (x in each field _c, y _c) the absolute difference i of gray-scale value:

i＝|f(u,v)-f(x _c,y _c)|＝|f(u,v)-g _c|

F (u, v) and f (x in formula _c, y _c)=g _crepresent pixel (u, v) and central pixel point (x respectively _c, y _c) gray-scale value, g _c=0,1 ..., M _g-1, i=0,1 ..., M _g-1;

Step 2: statistics absolute difference is the number of i

$s_{g_c,i}=\#\left\{(u,v)\right|(u,v)\&Element;V_D(x_c,y_c),i\}$

In formula, # represents in statistics set { (u, v) | () } number of the point (u, v) met under specified criteria ();

Step 3: the three-dimensional data table building each central pixel point of froth images;

Define certain central pixel point (x _c, y _c) three-dimensional data table f _d(x _c, y _c, i) as follows:

$f_D(x_c,y_c,i)=table(g_c,s_{g_c,i},i)$

In formula be that in a table, element is respectively g _c, i, size is 1 × 1 × M _gthree-dimensional data table;

For central pixel point all in froth images, build corresponding three-dimensional data table.

4. the compounded texture feature extracting method of floatation foam image according to claim 3, is characterized in that, the nested gray scale frequency table that described step 3 builds based on three-dimensional data table is specially:

Define nested gray scale frequency table F _d(x ' _c, y ' _c, i) as follows:

F _D(x _c',y _c',i)＝table(f _D(x _c,y _c,i))

X in formula _c'=x _c-D, x _c' ∈ 0,1 ..., k-1-2D; y _c'=y _c-D, y _c' ∈ 0,1 ..., m-1-2D, nested gray scale frequency table F _d(x ' _c, y ' _c, certain cell location (x i) _c', y _c') element be

By step 2 for all central pixel point (x _c, y _c) constructed by each three-dimensional data table f _d(x _c, y _c, i), embed respective cells position in nested gray scale frequency table, construct a size for (k-2D) × (m-2D) × M _gnested gray scale frequency table.

5. the compounded texture feature extracting method of floatation foam image according to claim 4, is characterized in that, the neighboring gray level dependence matrix that described step 4 obtains a kind of improvement is specially:

Definition improves neighborhood correlation matrix Q _d(g, i) is as follows:

G=0 in formula, 1 ..., M _g-1, i=0,1 ..., M _g-1, q _d(g, i) is matrix Q _d(g, i) is at the element at position (g, i) place;

At the nested gray scale frequency table F that step 3 obtains _d(x ' _c, y ' _c, i), statistics neighborhood territory pixel point gray-scale value and central pixel point gray-scale value g _cabsolute difference all when being i summation q _d(g _c, i):

In formula, Σ represents set { s _g,i| () } in meet all s of specified criteria () _g,Isum, & & presentation logic and computing;

By q _d(g _c, i) as matrix Q _d(g, i) is at position (g=g _c, the i) element at place, obtains a M _g× M _gimprovement neighboring gray level dependence matrix Q _d(g, i).

6. the compounded texture feature extracting method of floatation foam image according to claim 5, is characterized in that, described step 5 calculates a kind of new compounded texture feature and is specially:

According to the neighboring gray level dependence matrix improved, define a kind of new secondary statistical nature, namely compounded texture feature CT is:

$CT=\frac{\underset{g=0}{\overset{M_g-1}{\&Sigma;}}{g}^2\left(\underset{i=0}{\overset{M_g-1}{\&Sigma;}}{Q}_D\right(g,i\left)\right)}{\underset{g=0}{\overset{M_g-1}{\&Sigma;}}\underset{i=0}{\overset{M_g-1}{\&Sigma;}}\&lsqb;{(i+1)}^2{Q}_D\left(g,i\right)\&rsqb;}.$