CN105160651A - Paper defect detection method based on vision attention mechanism - Google Patents

Abstract

The invention relates o a paper defect detection method based on the vision attention mechanism and belongs to the papermaking technology field. The paper defect detection method based on the vision attention mechanism is provided through researching the vision attention mechanism. According to the paper defect detection method, an image with a paper defect can be easily discovered through eye observation as the paper defect more easily attracts eye attention compared with background textures, through simulating the vision attention mechanism, a paper defect detection calculation model based on the vision attention mechanism is established. Compared with a traditional paper detect detection method, the paper defect detection method provided in the invention has properties of strong detection robustness and rapid speed, needs no threshold calculation, adaptively carries out paper defect area calculation and can satisfy paper detection requirements in different background conditions, such as textile industry and cigarette paper industry.

Description

The paper defect testing method of view-based access control model attention mechanism

Technical field

The present invention relates to the paper defect testing method of view-based access control model attention mechanism, belong to pulp technology for making paper.

Background technology

In paper-making process, the appearance of paper defects has a strong impact on the judge of people to paper quality and the benefit in paper mill, therefore, fast, accurately, stably judges that the production of the type of paper defects to modern paper industry is significant.Paper defect testing method relatively more conventional is at present threshold determination method, found by research, its deficiency of threshold determination method is that different paper needs to choose different gray thresholds, and for the paper defects of faint paper defects, special shape paper defects and low contrast, adopt threshold determination method to be difficult to obtain satisfied effect.

The present invention is by studying human eye vision attention mechanism, propose a kind of paper defect testing method of new view-based access control model attention mechanism, eye-observation one width contains the image of paper defects, can be easy to find out, and its reason is paper defects is the most easily attract human eye to note for background texture.

Summary of the invention

Based on this, the present invention devises the paper defect testing method of view-based access control model attention mechanism.

The technical scheme that the present invention takes is:

The paper defect testing method of view-based access control model attention mechanism, concrete steps are:

The first step:

Multi-scale sampling and linear filtering, multi-scale sampling is from the original input picture bottom, from down to up, each tomographic image is that original image is 0 yardstick, then often sampling should be carried out, and yardstick adds 1 by obtaining its adjacent lower image sampling successively, the resolution of image along with yardstick increase with 2 the factor successively decrease, be defined as totally 5 yardsticks in computation process mesoscale, the present invention adopts Gaussian wave filter to carry out filtering to employing image, makes { x _ijrepresent original image, x _ijrepresent the gray-scale value at coordinate (i, j) place in image, then adopt formula to carry out the calculating of class Gaussian convolution to the every bit in image

$x_{ij}^{\left(0\right)}=x_{ij}$

$x_{ij}^{\left(1\right)}=\underset{p=-2}{\overset{p=2}{\Σ}}\underset{q=-2}{\overset{q=2}{\Σ}}{g}_{pq}{x}_{i-p,j-q}^{\left(0\right)}$

…

$x_{ij}^{\left(\mathrm{\σ}\right)}=\underset{p=-2}{\overset{p=2}{\Σ}}\underset{q=-2}{\overset{q=2}{\Σ}}{g}_{pq}{x}_{i-p,j-q}^{(\mathrm{\σ}-1)}$

Wherein, convolution matrix for:

$\[g_{pq}\]=\left[\begin{array}{ccccc}1& 4& 6& 4& 1\\ 4& 16& 24& 16& 4\\ 6& 24& 36& 24& 6\\ 4& 16& 24& 16& 4\\ 1& 4& 6& 4& 1\end{array}\right],(p,q=-2,-1,0,1,2).$

Second step:

Extract brightness, color and towards characteristic pattern respectively, brightness figure FM _i(σ)=(r (σ)+g (σ)+b (σ))/3 wherein, FM _i(σ) brightness figure under σ yardstick is represented, r (σ), g (σ), b (σ) represent the red, green, blue triple channel information of input picture under σ yardstick respectively, red-green (Red-Green of color characteristic figure, RG) and yellow-blue (Yellow-Blue, BY) fly up and down to representing its feature;

${\mathrm{FM}}_C^{rg}\left(\mathrm{\σ}\right)=\frac{r\left(\mathrm{\σ}\right)-g\left(\mathrm{\σ}\right)}{max\left(r\right(\mathrm{\σ}),g(\mathrm{\σ}),b(\mathrm{\σ}\left)\right)}$

${\mathrm{FM}}_C^{by}\left(\mathrm{\σ}\right)=\frac{b\left(\mathrm{\σ}\right)-min\left(r\right(\mathrm{\σ}),g(\mathrm{\σ}\left)\right)}{max\left(r\right(\mathrm{\σ}),g(\mathrm{\σ}),b(\mathrm{\σ}\left)\right)}$

Towards characteristic pattern ${\mathrm{FM}}_O^{\mathrm{\θ}}\left(\mathrm{\σ}\right)=\left|\right|F_I\left(\mathrm{\σ}\right)*G_0\left(\mathrm{\θ}\right)\left|\right|+\left|\right|F_I\left(\mathrm{\σ}\right)*G_{\mathrm{\π}/2}\left(\mathrm{\θ}\right)\left|\right|,$ Wherein, for Gabor function, θ ∈ 0 °, and 45 °, 90 °, 135 ° }, x'=xcos (θ)-ysin (θ); Y'=-xsin (θ)-ycos (θ)

The present invention chooses γ=1, λ=7, wave filter is taken as the matrix of 19 × 19, δ represents the reach of receptive field, δ can obtain different Gabor templates when getting different values, the effect of cosine function can be strengthened when δ is too large, Gaussian function effect is not obvious, the effect of each pixel in filtering is nearly all the same, when δ is too little, filter action only shows the zone line of template, field point works hardly in filtering, when only having δ value moderate Gabor function competence exertion its obtain towards effect, compare the present invention through experiment and choose δ=3.5.

3rd step:

Extract brightness, color and towards Characteristic Contrast mapping graph respectively, Characteristic Contrast mapping graph is obtained by doing difference to the characteristic pattern under different scale, first, the characteristic pattern of different scale is become the information under same yardstick by interpolation or extraction, then, carry out point-to-point subtraction, if center is c, periphery is s, note central authorities-periphery difference be operating as Θ, then brightness, color and towards contrast mapping graph can try to achieve respectively:

CM _I(c,s)＝|FM _I(c)ΘFM _I(s)|

${\mathrm{CM}}_C^{rg}(c,s)=\left|{\mathrm{FM}}_C^{rg}\left(c\right){\mathrm{\ΘFM}}_C^{rg}\left(s\right)\right|$

${\mathrm{CM}}_C^{by}(c,s)=\left|{\mathrm{FM}}_C^{by}\left(c\right){\mathrm{\ΘFM}}_C^{by}\left(s\right)\right|$

${\mathrm{CM}}_O^{\mathrm{\θ}}(c,s)=\left|{\mathrm{FM}}_O^{\mathrm{\θ}}\left(c\right){\mathrm{\ΘFM}}_O^{\mathrm{\θ}}\left(s\right)\right|.$

4th step:

Obtain brightness, color and significantly scheming towards son, to all characteristic informations, if the contrast mapping graph directly merged under all yardsticks obtains son significantly figure, then obtained son remarkable figure feature may weaken self more significant information due to the impact of noise, therefore, first the present invention was normalized contrast mapping graph before acquisition significantly figure, normalization is represented with " N () ", comprise three steps: the information of contrast mapping graph is normalized to [0 ... K] in scope, object is that the feature extracting method eliminated due to different characteristic is different, and the problem that the maximal value of each characteristic remarkable picture caused is different, to each width characteristic pattern, in computed image, remove the mean value of all local maximal point of the remainder outside global maximum point K calculate the value obtained is exactly the weighting merge coefficient of this characteristic remarkable picture, can obtain brightness, color respectively and significantly scheme towards son according to above step,

$\stackrel{\‾}{C}{M}_I=\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_I\right(c,s\left)\right)$

$\stackrel{\‾}{C}{M}_C=\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_C^{rg}\right(c,s\left)\right)+\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_C^{by}\right(c,s\left)\right)$

$\stackrel{\‾}{C}{M}_O=\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_O\right(c,s\left)\right),$ θ∈{0°，45°，90°，135°}。

5th step:

Obtain the overall situation significantly figure, in the same way the remarkable figure of son of brightness, color and orientation information is normalized, then carries out merging and obtain the overall situation significantly figure.

6th step:

Paper defects region calculates, paper defects region is corresponding in remarkable figure is exactly marking area in the remarkable figure calculated by institute of the present invention extracting method, the present invention produces several marking area by the competition of conspicuousness power, competition mechanism that relatively conventional competition mechanism is that the victor is a king (Winner-take-all), namely the several marking areas in remarkable figure compare, saliency value is larger, first the region that conspicuousness is stronger obtains the attention of human eye, once after noticing that target is determined, other parts in scene can not be reentried attention, forbid that return mechanisms (InhibitionofReturn) is another the important mechanism in marking area transfer, in the process finding marking area, the marking area noted will no longer participate in the transfer process of marking area, namely each marking area is for once by the chance noted, by the victor is a king and forbid that return mechanisms obtains paper defects region.

Native system has following characteristics:

1., by simulation human eye vision attention mechanism, set up the paper defect testing computation model of view-based access control model attention mechanism.

2., compared with traditional paper defects detection method, method provided by the present invention detects strong robustness, speed is fast, does not need to carry out threshold calculations.

3. self-adaptation carries out the calculating of paper defects region, can meet the paper defect testing demand under different background condition, as textile industry, cigarette paper industry etc.

More than show and describe ultimate principle of the present invention and principal character and advantage of the present invention.The technician of the industry should understand; the present invention is not restricted to the described embodiments; what describe in above-described embodiment and instructions just illustrates principle of the present invention; without departing from the spirit and scope of the present invention; the present invention also has various changes and modifications, and these changes and improvements all fall in the claimed scope of the invention.Application claims protection domain is defined by appending claims and equivalent thereof.

Accompanying drawing explanation

Fig. 1 is web inspection system block diagram of the present invention.

Embodiment

Below in conjunction with specific embodiment, the invention will be further described, is used for explaining the present invention in this illustrative examples of inventing and explanation, but not as a limitation of the invention.

As shown in Figure 1, the paper defect testing method of view-based access control model attention mechanism, concrete steps are:

The first step:

Multi-scale sampling and linear filtering, multi-scale sampling is from the original input picture bottom, from down to up, each tomographic image is that original image is 0 yardstick, then often sampling should be carried out, and yardstick adds 1 by obtaining its adjacent lower image sampling successively, the resolution of image along with yardstick increase with 2 the factor successively decrease, be defined as totally 5 yardsticks in computation process mesoscale, the present invention adopts Gaussian wave filter to carry out filtering to employing image, makes { x _ijrepresent original image, x _ijrepresent the gray-scale value at coordinate (i, j) place in image, then adopt formula (1) to carry out the calculating of class Gaussian convolution to the every bit in image

$x_{ij}^{\left(0\right)}=x_{ij}$

$x_{ij}^{\left(1\right)}=\underset{p=-2}{\overset{p=2}{\Σ}}\underset{q=-2}{\overset{q=2}{\Σ}}{g}_{pq}{x}_{i-p,j-q}^{\left(0\right)}$

…

$x_{ij}^{\left(\mathrm{\σ}\right)}=\underset{p=-2}{\overset{p=2}{\Σ}}\underset{q=-2}{\overset{q=2}{\Σ}}{g}_{pq}{x}_{i-p,j-q}^{(\mathrm{\σ}-1)}$

Wherein, convolution matrix for:

$\[g_{pq}\]=\left[\begin{array}{ccccc}1& 4& 6& 4& 1\\ 4& 16& 24& 16& 4\\ 6& 24& 36& 24& 6\\ 4& 16& 24& 16& 4\\ 1& 4& 6& 4& 1\end{array}\right],(p,q=-2,-1,0,1,2).$

Second step:

Extract brightness, color and towards characteristic pattern respectively, brightness figure FM _i(σ)=(r (σ)+g (σ)+b (σ))/3 wherein, FM _i(σ) brightness figure under σ yardstick is represented, r (σ), g (σ), b (σ) represent the red, green, blue triple channel information of input picture under σ yardstick respectively, red-green (Red-Green of color characteristic figure, RG) and yellow-blue (Yellow-Blue, BY) fly up and down to representing its feature;

${\mathrm{FM}}_C^{rg}\left(\mathrm{\σ}\right)=\frac{r\left(\mathrm{\σ}\right)-g\left(\mathrm{\σ}\right)}{max\left(r\right(\mathrm{\σ}),g(\mathrm{\σ}),b(\mathrm{\σ}\left)\right)}$

${\mathrm{FM}}_C^{by}\left(\mathrm{\σ}\right)=\frac{b\left(\mathrm{\σ}\right)-min\left(r\right(\mathrm{\σ}),g(\mathrm{\σ}\left)\right)}{max\left(r\right(\mathrm{\σ}),g(\mathrm{\σ}),b(\mathrm{\σ}\left)\right)}$

Towards characteristic pattern ${\mathrm{FM}}_O^{\mathrm{\θ}}\left(\mathrm{\σ}\right)=\left|\right|F_I\left(\mathrm{\σ}\right)*G_0\left(\mathrm{\θ}\right)\left|\right|+\left|\right|F_I\left(\mathrm{\σ}\right)*G_{\mathrm{\π}/2}\left(\mathrm{\θ}\right)\left|\right|,$ Wherein, for Gabor function, θ ∈ 0 °, and 45 °, 90 °, 135 ° }, x'=xcos (θ)-ysin (θ); Y'=-xsin (θ)-ycos (θ)

The present invention chooses γ=1, λ=7, wave filter is taken as the matrix of 19 × 19, δ represents the reach of receptive field, δ can obtain different Gabor templates when getting different values, the effect of cosine function can be strengthened when δ is too large, Gaussian function effect is not obvious, the effect of each pixel in filtering is nearly all the same, when δ is too little, filter action only shows the zone line of template, field point works hardly in filtering, when only having δ value moderate Gabor function competence exertion its obtain towards effect, compare the present invention through experiment and choose δ=3.5.

3rd step:

Extract brightness, color and towards Characteristic Contrast mapping graph respectively, Characteristic Contrast mapping graph is obtained by doing difference to the characteristic pattern under different scale, first, the characteristic pattern of different scale is become the information under same yardstick by interpolation or extraction, then, carry out point-to-point subtraction, if center is c, periphery is s, note central authorities-periphery difference be operating as Θ, then brightness, color and towards contrast mapping graph can be tried to achieve by formula (5) ~ (7) respectively:

CM _I(c,s)＝|FM _I(c)ΘFM _I(s)|

${\mathrm{CM}}_C^{rg}(c,s)=\left|{\mathrm{FM}}_C^{rg}\left(c\right){\mathrm{\ΘFM}}_C^{rg}\left(s\right)\right|$ ${\mathrm{CM}}_C^{by}(c,s)=\left|{\mathrm{FM}}_C^{by}\left(c\right){\mathrm{\ΘFM}}_C^{by}\left(s\right)\right|$

${\mathrm{CM}}_O^{\mathrm{\θ}}(c,s)=\left|{\mathrm{FM}}_O^{\mathrm{\θ}}\left(c\right){\mathrm{\ΘFM}}_O^{\mathrm{\θ}}\left(s\right)\right|.$

4th step:

Obtain brightness, color and significantly scheming towards son, to all characteristic informations, if the contrast mapping graph directly merged under all yardsticks obtains son significantly figure, then obtained son remarkable figure feature may weaken self more significant information due to the impact of noise, therefore, first the present invention was normalized contrast mapping graph before acquisition significantly figure, normalization is represented with " N () ", comprise three steps: the information of contrast mapping graph is normalized to [0 ... K] in scope, object is that the feature extracting method eliminated due to different characteristic is different, and the problem that the maximal value of each characteristic remarkable picture caused is different, to each width characteristic pattern, in computed image, remove the mean value of all local maximal point of the remainder outside global maximum point K calculate the value obtained is exactly the weighting merge coefficient of this characteristic remarkable picture, can obtain brightness, color respectively and significantly scheme towards son according to above step,

$\stackrel{\‾}{C}{M}_I=\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_I\right(c,s\left)\right)$

$\stackrel{\‾}{C}{M}_C=\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_C^{rg}\right(c,s\left)\right)+\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_C^{by}\right(c,s\left)\right)$

$\stackrel{\‾}{C}{M}_O=\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_O\right(c,s\left)\right),$ θ∈{0°，45°，90°，135°}。

5th step:

Obtain the overall situation significantly figure, in the same way the remarkable figure of son of brightness, color and orientation information is normalized, then carries out merging and obtain the overall situation significantly figure.

6th step:

Paper defects region calculates, paper defects region is corresponding in remarkable figure is exactly marking area in the remarkable figure calculated by institute of the present invention extracting method, the present invention produces several marking area by the competition of conspicuousness power, competition mechanism that relatively conventional competition mechanism is that the victor is a king (Winner-take-all), namely the several marking areas in remarkable figure compare, saliency value is larger, first the region that conspicuousness is stronger obtains the attention of human eye, once after noticing that target is determined, other parts in scene can not be reentried attention, forbid that return mechanisms (InhibitionofReturn) is another the important mechanism in marking area transfer, in the process finding marking area, the marking area noted will no longer participate in the transfer process of marking area, namely each marking area is for once by the chance noted, by the victor is a king and forbid that return mechanisms obtains paper defects region.

Claims (1)

1. the paper defect testing method of view-based access control model attention mechanism, is characterized in that: concrete steps are:

The first step:

Multi-scale sampling and linear filtering, multi-scale sampling is from the original input picture bottom, from down to up, each tomographic image is that original image is 0 yardstick, then often sampling should be carried out, and yardstick adds 1 by obtaining its adjacent lower image sampling successively, the resolution of image along with yardstick increase with 2 the factor successively decrease, be defined as totally 5 yardsticks in computation process mesoscale, the present invention adopts Gaussian wave filter to carry out filtering to employing image, makes { x _ijrepresent original image, x _ijrepresent the gray-scale value at coordinate (i, j) place in image, then adopt formula to carry out the calculating of class Gaussian convolution to the every bit in image

$x_{ij}^{\left(0\right)}=x_{ij}$

$x_{ij}^{\left(1\right)}=\underset{p=-2}{\overset{p=2}{\Σ}}\underset{q=-2}{\overset{q=2}{\Σ}}{g}_{pq}{x}_{i-p,j-q}^{\left(0\right)}$

…

$x_{ij}^{\left(\mathrm{\σ}\right)}=\underset{p=-2}{\overset{p=2}{\Σ}}\underset{q=-2}{\overset{q=2}{\Σ}}{g}_{pq}{x}_{i-p,j-q}^{(\mathrm{\σ}-1)}$

Wherein, convolution matrix [g _pq] be:

$\[g_{pq}\]=\left[\begin{array}{ccccc}1& 4& 6& 4& 1\\ 4& 16& 24& 16& 4\\ 6& 24& 36& 24& 6\\ 4& 16& 24& 16& 4\\ 1& 4& 6& 4& 1\end{array}\right],(p,q=-2,-1,0,1,2);$

Second step:

Extract brightness, color and towards characteristic pattern respectively, brightness figure FM _i(σ)=(r (σ)+g (σ)+b (σ))/3 wherein, FM _i(σ) brightness figure under σ yardstick is represented, r (σ), g (σ), b (σ) represent the red, green, blue triple channel information of input picture under σ yardstick respectively, red-green (Red-Green of color characteristic figure, RG) and yellow-blue (Yellow-Blue, BY) fly up and down to representing its feature;

${\mathrm{FM}}_C^{rg}\left(\mathrm{\σ}\right)=\frac{r\left(\mathrm{\σ}\right)-g\left(\mathrm{\σ}\right)}{max\left(r\right(\mathrm{\σ}),g(\mathrm{\σ}),b(\mathrm{\σ}\left)\right)}$

${\mathrm{FM}}_C^{by}\left(\mathrm{\σ}\right)=\frac{b\left(\mathrm{\σ}\right)-min\left(r\right(\mathrm{\σ}),g(\mathrm{\σ}\left)\right)}{max\left(r\right(\mathrm{\σ}),g(\mathrm{\σ}),b(\mathrm{\σ}\left)\right)}$

Towards characteristic pattern ${\mathrm{FM}}_O^{\mathrm{\θ}}\left(\mathrm{\σ}\right)=\left|\right|F_I\left(\mathrm{\σ}\right)*G_0\left(\mathrm{\θ}\right)\left|\right|+\left|\right|F_I\left(\mathrm{\σ}\right)*G_{\mathrm{\π}/2}\left(\mathrm{\θ}\right)\left|\right|,$ Wherein, for Gabor function, θ ∈ 0 °, and 45 °, 90 °, 135 ° },

$G_{\mathrm{\φ}}(x,y,\mathrm{\θ})=\exp (-\frac{x^{\′2}+{\mathrm{\γ}}^2{y}^{\′2}}{2{\mathrm{\δ}}^2})cos(2\mathrm{\π}\frac{x^{\′}}{\mathrm{\λ}}+\mathrm{\ψ});$ x'＝xcos(θ)-ysin(θ)；y'＝-xsin(θ)-ycos(θ)

The present invention chooses γ=1, λ=7, wave filter is taken as the matrix of 19 × 19, δ represents the reach of receptive field, δ can obtain different Gabor templates when getting different values, the effect of cosine function can be strengthened when δ is too large, Gaussian function effect is not obvious, the effect of each pixel in filtering is nearly all the same, when δ is too little, filter action only shows the zone line of template, field point works hardly in filtering, when only having δ value moderate Gabor function competence exertion its obtain towards effect, compare the present invention through experiment and choose δ=3.5,

3rd step:

Extract brightness, color and towards Characteristic Contrast mapping graph respectively, Characteristic Contrast mapping graph is obtained by doing difference to the characteristic pattern under different scale, first, the characteristic pattern of different scale is become the information under same yardstick by interpolation or extraction, then, carry out point-to-point subtraction, if center is c, periphery is s, note central authorities-periphery difference be operating as Θ, then brightness, color and towards contrast mapping graph can try to achieve respectively:

CM _I(c,s)＝|FM _I(c)ΘFM _I(s)|

${\mathrm{CM}}_C^{\mathrm{rg}}(c,s)=\left|{\mathrm{FM}}_C^{\mathrm{rg}}\left(c\right)\mathrm{\Θ}{\mathrm{FM}}_C^{\mathrm{rg}}\left(s\right)\right|$

${\mathrm{CM}}_C^{by}(c,s)=\left|{\mathrm{FM}}_C^{by}\left(c\right){\mathrm{\ΘFM}}_C^{by}\left(s\right)\right|$

${\mathrm{CM}}_O^{\mathrm{\θ}}(c,s)=\left|{\mathrm{FM}}_O^{\mathrm{\θ}}\left(c\right){\mathrm{\ΘFM}}_O^{\mathrm{\θ}}\left(s\right)\right|;$

4th step:

Obtain brightness, color and significantly scheming towards son, to all characteristic informations, if the contrast mapping graph directly merged under all yardsticks obtains son significantly figure, then obtained son remarkable figure feature may weaken self more significant information due to the impact of noise, therefore, first the present invention was normalized contrast mapping graph before acquisition significantly figure, normalization is represented with " N () ", comprise three steps: the information of contrast mapping graph is normalized to [0 ... K] in scope, object is that the feature extracting method eliminated due to different characteristic is different, and the problem that the maximal value of each characteristic remarkable picture caused is different, to each width characteristic pattern, in computed image, remove the mean value of all local maximal point of the remainder outside global maximum point K calculate the value obtained is exactly the weighting merge coefficient of this characteristic remarkable picture, can obtain brightness, color respectively and significantly scheme towards son according to above step,

$\stackrel{\‾}{C}{M}_I=\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_I\right(c,s\left)\right)$

$\stackrel{\‾}{C}{M}_C=\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_c^{rg}\right(c,s\left)\right)+\underset{c\∈\{0,1\},s\∈\{2,3,4\}}{\Σ}N\left({\mathrm{CM}}_C^{by}\right(c,s\left)\right)$

5th step:

Obtain the overall situation significantly figure, in the same way the remarkable figure of son of brightness, color and orientation information is normalized, then carries out merging and obtain the overall situation significantly figure;

6th step:

Paper defects region calculates, paper defects region is corresponding in remarkable figure is exactly marking area in the remarkable figure calculated by institute of the present invention extracting method, the present invention produces several marking area by the competition of conspicuousness power, competition mechanism that relatively conventional competition mechanism is that the victor is a king (Winner-take-all), namely the several marking areas in remarkable figure compare, saliency value is larger, first the region that conspicuousness is stronger obtains the attention of human eye, once after noticing that target is determined, other parts in scene can not be reentried attention, forbid that return mechanisms (InhibitionofReturn) is another the important mechanism in marking area transfer, in the process finding marking area, the marking area noted will no longer participate in the transfer process of marking area, namely each marking area is for once by the chance noted, by the victor is a king and forbid that return mechanisms obtains paper defects region.