CN103106672B - A kind of interesting image regions detection method based on color characteristic - Google Patents

Abstract

The present invention relates to a kind of interesting image regions detection method based on color characteristic, it is characterized in that: first, consider Lab and antagonistic pairs two color spaces simultaneously, consider each pixel and multiple dimensioned neighborhood difference problem simultaneously, so not only consider the global feature of image but also consider the local feature of image.Using information entropy as weighing the Detection results of remarkable figure, first using information entropy to find optimum three neighborhood yardsticks in each Color Channel, then finding optimum three Color Channels by information entropy, and the combination of operation information entropy often walks the optimum obtained significantly schemes.So not only consider the global feature of image but also consider the local feature of image.Then information entropy is used to combine the optimum conspicuousness distribution plan obtained step by step.

Description

A kind of interesting image regions detection method based on color characteristic

Technical field

The invention belongs to digital image processing field, be specifically related to a kind of interesting image regions detection method based on color characteristic.

Background technology

Along with the development of infotech, the especially innovation day by day of electronic product, digital camera is popularized in daily life greatly, and the acquisition of image is more and more easier.New problem served by the rapid growth band of view data, and as compression of images, transmission, stores, browse, retrieval, tissue and Image mining etc.This demand proposes more and more higher requirement to computer information processing efficiency.In human visual attention mechanism, the content that task is concerned about is only a part very little in entire image usually.All data being put on an equal footing is do not meet human vision characteristics.In computer picture information processing, how to simulate and to realize the Selective Attention Mechanism of the mankind, the image-region finding those easily to cause observer fast to note, form a rational computational resource allocation scheme, guide whole Image Information Processing process, make computing machine have the Image Information Processing ability of the anthropoid selectivity of class and initiative, will be significant to raising computer picture information processing efficiency.

Color is the most important factor affecting human visual system, Lab color space is and device-independent color model, closest to the vision of the mankind, therefore first we consider Lab color space, we consider again antagonistic pairs color space simultaneously, and red green antagonistic pairs and blue yellow antagonistic pairs more can embody the color distortion of image.We consider each pixel and multiple dimensioned neighborhood difference problem, so not only consider the global feature of image but also consider the local feature of image.Multiple Color Channel obtains multiple characteristic pattern, and the fusion of characteristic pattern is the steps necessary of conspicuousness research, the characteristic pattern choosing all passages can increase calculated amount, and the bad remarkable figure of effect can reduce the Detection job of final area-of-interest, therefore, we choose a part of effect preferably significantly figure merge and arrive final significantly figure.

Summary of the invention

The technical matters solved

In order to avoid the deficiencies in the prior art part, the present invention proposes a kind of interesting image regions detection method based on color characteristic, detects the interesting target meeting human vision characteristics in image fast and effectively.

Technical scheme

Based on an interesting image regions detection method for color characteristic, it is characterized in that step is as follows:

Step 1: be four look wideband color space RGBY by RGB color space conversion by original input picture, method for transformation is:

$R=r-\frac{g+b}{2}$

$G=g-\frac{r+b}{2}$

$B=b-\frac{r+g}{2}$

$Y=\frac{r+g}{2}-\left|\frac{r-g}{2}\right|-b$

Wherein: r, g, b represent the red, green, blue Color Channel in RGB color space respectively, R, G, B, Y represent the red, green, blue obtained in RGBY color space, yellow Color Channel respectively;

Step 2: calculate brightness I=(r+g+b), red green antagonistic pairs RG=|R-G|, blue yellow antagonistic pairs BY=|B-Y|, obtain brightness, red green antagonistic pairs and blue yellow antagonistic pairs three passage I, RG, BY respectively;

Step 3: original input picture is converted into Lab color space, obtains L, a, b Color Channel;

Step 4: calculate the pixel average mean_N in two Square Neighborhoods of each point (x, y) in 6 Color Channels _1i(x, y) and mean_N _2i(x, y), i ∈ { I, RG, BY, L, a, b};

Wherein: mean_N _1i(x, y) and mean_N _2ithe size that (x, y) is illustrated respectively in Color Channel i mid point (x, y) place is N ₁× N ₁and N ₂× N ₂square Neighborhood in pixel average;

Step 5: calculate 6 Color Channel significant characteristics:

C _i(x,y)＝|mean_N _1i(x,y)-mean_N _2i(x,y)|，i∈{I,RG,BY,L,a,b}

Wherein: C _i(x, y) represents the significant characteristics at Color Channel i mid point (x, y) place;

Step 6: successively at N ₁value N ₁₁, N ₁₂, successively at N ₂value N ₂₁, N ₂₂..., N _2m, repeat step 5 and can obtain 2m notable feature figure at each Color Channel;

Step 7: after characteristic pattern normalization, the information entropy calculating characteristic pattern obtains:

$H\left(C_{ij}\right(x,y))=-\underset{x=1}{\overset{n}{\Σ}}\underset{y=1}{\overset{n}{\Σ}}{\mathrm{gC}}_{ij}(x,y)log\left({\mathrm{gC}}_{ij}\right(x,y)),$ i∈{I,RG,BY,L,a,b}，j＝1,2,…,2m

Wherein: C _ij(x, y) is that i Color Channel jth is significantly schemed, and g is a gauss low frequency filter;

Step 8: at 2m notable feature figure of each Color Channel, gets three notable feature figure Optimalmap that information entropy is minimum respectively _i1, Optimalmap _i2, Optimalmap _i3, i ∈ { I, RG, BY, L, a, b};

Step 9: by Optimalmap _i1, Optimalmap _i2and Optimalmap _i3normalization is also combined, and obtains i Color Channel significant characteristics:

${\mathrm{Smap}}_i=\underset{j=1}{\overset{3}{\Σ}}{w}_{ij}{\mathrm{Optimalmap}}_{ij},$

Wherein: $w_{ij}=\frac{1}{H\left({\mathrm{Optimalmap}}_{ij}\right)\×\underset{j=1}{\overset{3}{\Σ}}\frac{1}{H\left({\mathrm{Optimalmap}}_{ij}\right)}}$ For Fusion Features weight coefficient;

Step 10: the method repeating step 8 and 9, obtains three Color Channels that information entropy is minimum, and obtains the final significantly figure of interesting image regions to normalization combination.

Beneficial effect

A kind of interesting image regions detection method based on color characteristic that the present invention proposes, first, consider Lab and antagonistic pairs two color spaces simultaneously, consider each pixel and multiple dimensioned neighborhood difference problem simultaneously, so not only consider the global feature of image but also consider the local feature of image.Using information entropy as weighing the Detection results of remarkable figure, first using information entropy to find optimum three neighborhood yardsticks in each Color Channel, then finding optimum three Color Channels by information entropy, and the combination of operation information entropy often walks the optimum obtained significantly schemes.So not only consider the global feature of image but also consider the local feature of image.Then information entropy is used to combine the optimum conspicuousness distribution plan obtained step by step.

Accompanying drawing explanation

Fig. 1: the basic flow sheet of the inventive method

Embodiment

Now in conjunction with the embodiments, the invention will be further described for accompanying drawing:

Hardware environment for implementing is: AMDAthlon64 × 25000+ computing machine, 2GB internal memory, 256M video card, the software environment of operation is: Matlab2010a and WindowsXP.The method that we use the present invention of Matlab software simulating to propose.

The present invention is specifically implemented as follows:

1: the size of input picture is adjusted to 128 × 128;

2: input picture is converted into Lab color space, obtain three Color Channel L, a, b;

3: image is converted into rgb color space, r, g, b represent RGB three Color Channels respectively, calculate

$R=r-\frac{g+b}{2}$

$G=g-\frac{r+b}{2}$

$B=b-\frac{r+g}{2}$

$Y=\frac{r+g}{2}-\left|\frac{r-g}{2}\right|-b$

Obtain RGBY color space;

4: calculate brightness I=(r+g+b), calculate antagonistic pairs RG=|R-G|, BY=|B-Y|, obtain other three passage I, RG, BY;

5: calculate the pixel average in two Square Neighborhoods of each point (x, y) in 6 Color Channels, the size of two neighborhoods is N ₁× N ₁and N ₂× N ₂;

6: calculate 6 Color Channel significant characteristics

C(x,y)＝|mean_N ₁-mean_N ₂|，

Wherein: mean_N ₁and mean_N ₂represent that the size at point (x, y) place is N respectively ₁× N ₁and N ₂× N ₂square Neighborhood in pixel average,

7:N ₁value 3 and 5, N successively ₂value 128,64,32,16 successively, uses each Color Channel of step 6 can obtain 8 notable feature figure, is total to 48 are significantly schemed;

8: after characteristic pattern normalization, the information entropy calculating characteristic pattern obtains

$H\left(C_{ij}\right(x,y))=-\underset{x=1}{\overset{n}{\Σ}}\underset{y=1}{\overset{n}{\Σ}}{\mathrm{gC}}_{ij}(x,y)log\left({\mathrm{gC}}_{ij}\right(x,y)),$ i∈{I,RG,BY,L,a,b}，j＝1,2,…,8

Wherein: C _ij(x, y) is that i Color Channel jth is significantly schemed, and g is a gauss low frequency filter;

9: in 8 notable feature figure of each Color Channel, get three notable feature figure Optimalmap that information entropy is minimum respectively _i1, Optimalmap _i2, Optimalmap _i3, i ∈ { I, RG, BY, L, a, b};

10: by Optimalmap _i1, Optimalmap _i2and Optimalmap _i3normalization is also combined, and obtains i Color Channel significant characteristics

${\mathrm{Smap}}_i=\underset{i=1}{\overset{3}{\Σ}}{w}_{ij}{\mathrm{Optimalmap}}_{ij},$

Wherein:

$w_{ij}=\frac{1}{H\left({\mathrm{Optimalmap}}_{ij}\right)\×\underset{j=1}{\overset{3}{\Σ}}\frac{1}{H\left({\mathrm{Optimalmap}}_{ij}\right)}}$ For Fusion Features weight coefficient;

11: in the notable feature figure of 6 Color Channels, get three notable feature figure OptimalSmap that information entropy is minimum respectively ₁, OptimalSmap ₂, OptimalSmap ₃;

10: by OptimalSmap ₁, OptimalSmap ₂, OptimalSmap ₃normalization is also combined, and obtains final significantly figure

$Smap=\underset{i=1}{\overset{3}{\Σ}}{w}_{ij}{\mathrm{OptimalSmap}}_i,$

Wherein:

$w_{ij}=\frac{1}{H\left({\mathrm{OptimalSmap}}_i\right)\×\underset{j=1}{\overset{3}{\Σ}}\frac{1}{H\left({\mathrm{OptimalSmap}}_i\right)}}$ For Fusion Features weight coefficient.

Claims (1)

1., based on an interesting image regions detection method for color characteristic, it is characterized in that step is as follows:

Step 1: be four look wideband color space RGBY by RGB color space conversion by original input picture, method for transformation is:

，

，

，

，

Wherein: r, g, b represent the red, green, blue Color Channel in RGB color space respectively, R, G, B, Y represent the red, green, blue obtained in RGBY color space, yellow Color Channel respectively;

Step 2: calculate brightness I=(r+g+b), red green antagonistic pairs RG=|R-G|, blue yellow antagonistic pairs BY=|B-Y|, obtain brightness, red green antagonistic pairs and blue yellow antagonistic pairs three passage I, RG, BY respectively;

Step 3: original input picture is converted into Lab color space, obtains L, a, b Color Channel;

Step 4: calculate the pixel average mean_N in two Square Neighborhoods of each point (x, y) in 6 Color Channels _1i(x, y) and mean_N _2i(x, y), i ∈ { I, RG, BY, L, a, b};

Wherein: mean_N _1i(x, y) and mean_N _2ithe size that (x, y) is illustrated respectively in Color Channel i mid point (x, y) place is N ₁× N ₁and N ₂× N ₂square Neighborhood in pixel average;

Step 5: calculate 6 Color Channel significant characteristics:

C _i(x,y)＝|mean_N _1i(x,y)-mean_N _2i(x,y)|，i∈{I,RG,BY,L,a,b}，

Wherein: C _i(x, y) represents the significant characteristics at Color Channel i mid point (x, y) place;

Step 6: successively at N ₁value N ₁₁, N ₁₂, successively at N ₂value N ₂₁, N ₂₂..., N _2m, repeat step 5 and can obtain 2m notable feature figure at each Color Channel;

Step 7: after characteristic pattern normalization, the information entropy calculating characteristic pattern obtains:

i∈{I,RG,BY,L,a,b}，j＝1,2,…,2m，

Wherein: C _ij(x, y) is that i Color Channel jth is significantly schemed, and g is a gauss low frequency filter;

Step 8: at 2m notable feature figure of each Color Channel, gets three notable feature figure Optimalmap that information entropy is minimum respectively _i1, Optimalmap _i2, Optimalmap _i3, i ∈ { I, RG, BY, L, a, b};

Step 9: by Optimalmap _i1, Optimalmap _i2and Optimalmap _i3normalization is also combined, and obtains i Color Channel significant characteristics:

Wherein: for Fusion Features weight coefficient;

Step 10: the method repeating step 8 and 9, obtains three Color Channels that information entropy is minimum, and obtains the final significantly figure of interesting image regions to normalization combination.