CN105023253A - Visual underlying feature-based image enhancement method - Google Patents

Abstract

The invention relates to a visual underlying feature-based image enhancement method. The method includes the following steps that: 1, the visual underlying features of an image are extracted; 2, weight fusion is performed on the visual underlying features of the image, so that a computing saliency map can be formed; 3, correlation coefficient comparison is performed on an eye movement saliency map measured by an eye tracker and the computing saliency map, so that the optimal weight of the visual underlying features can be determined; and 4, proper image enhancement methods can be selected for different types of images according to weight. According to the visual underlying feature-based image enhancement method of the invention, the eye fixation point-based eye movement saliency map is combined with a traditional image enhancement method for the first time, and an optimized visual underlying feature-based Itti visual attention model is adopted to extract the computing feature saliency map of the image, and therefore, the visual quality of color images re-displayed on a digital display device can be improved; and important information in the images can be highlighted according to different types of images, and less information is weakened, and therefore, a final image enhancement effect can more accords with eye visual perception. The method has a certain application value in image enhancement.

Description

The image enchancing method of view-based access control model low-level image feature

Technical field

The present invention relates to a kind of digital image processing method, especially a kind of image enchancing method of view-based access control model low-level image feature.

Background technology

In actual life, digital picture can cause because of a variety of causes the situation occurring distortion on the display device.In order to avoid this drawback, one of image enhancement technique research emphasis becoming Digital Image Processing.Show according to research, complicated human visual system when processing image be priority processing those attract the region (human eye vision area-of-interest) of human eye vision attention most, instead of the global information of image to be analyzed.

As everyone knows, the method that traditional images strengthens is all carry out treatment and analysis based on global information, this does not meet the characteristic of human visual system, not only cause many to the calculating waste in secondary information process analysis, reduce the efficiency of image procossing, and its final image enhancement effects do not meet the perception of human eye vision.Therefore, in order to overcome the drawback that traditional images Enhancement Method is brought, a kind of novel colour-image reinforcing method based on human eye visual perception becomes the focus of research.This method has considered human-eye visual characteristic and a large amount of physical experiment conclusion at heart.

Summary of the invention

The object of the invention is the image enchancing method proposing a kind of view-based access control model low-level image feature, the image enhancement effects making it final more meets human eye visual perception.

To achieve these goals, the technical solution used in the present invention is as follows:

An image enchancing method for view-based access control model low-level image feature, the method has following steps:

(1) extract the vision low-level image feature of image, comprise color, brightness, direction, texture and edge feature;

(2) Weighted Fusion becomes to calculate significantly figure;

(3) eye recorded with eye tracker moves remarkable figure to carry out related coefficient and compares, to determine the optimal weight of each vision low-level image feature;

(4) be that dissimilar image selects suitable image enchancing method according to the size of weight.

Described step (1) comprises following concrete steps further:

(1.1) in hsv color space, extract brightness, source images is changed into the gray-scale map only having monochrome information, finally image pixel value is normalized to again the scope of [0,255];

(1.2) under the hsv color space meeting human eye visual perception, extract color characteristic, after having extracted color characteristic, pixel value is normalized to again the scope of [0,255];

(1.3) direction character of image is extracted by Gabor transformation;

(1.4) Gabor filter group texture feature extraction is adopted: first utilize Gabor filter group to extract textural characteristics on 5 yardsticks of image, 4 directions, then this 20 width textural characteristics figure is normalized, finally all images are superimposed by equal weight, form last texture and significantly scheme;

(1.5) the most typical extracting mode of edge feature is the Laplace operator utilizing second-order differential, extracts the gray-scale edges feature of image in HSV space.

Described step (2) comprises following concrete steps further:

(2.1) Gaussian function is adopted respectively the color of input picture, brightness, direction, texture and edge to be carried out to the filtering of 6 yardsticks;

(2.2) Image Visual Feature under utilizing difference operation and Laplace operator to extract different scale;

(2.3) carry out central peripheral difference respectively to calculate, error image normalization is sued for peace, obtains the remarkable figure of this feature;

(2.4) carry out weight configuration for the 5 width characteristic remarkable pictures obtained, and ensure that the weighted value addition of 5 vision low-level image features equals 1; 5 width characteristic remarkable pictures are superposed, forms the remarkable figure after 1 width combination of multiple features.

The algorithmic formula that described step (3) adopts is as follows:

$\mathrm{\ρ}=\frac{{\mathrm{\Σ}}_x\left(H\right(x)-{\mathrm{\μ}}_H)\·\left(S\right(x)-{\mathrm{\μ}}_s)}{\sqrt{{\mathrm{\Σ}}_x{\left(H\right(x)-{\mathrm{\μ}}_H)}^2.{\mathrm{\Σ}}_x{\left(S\right(x)-{\mathrm{\μ}}_S)}^2}}$

In formula, the pixel in x representative image, H (x), S (x) are that eye moves the saliency value significantly scheming and calculate current pixel in remarkable figure, μ respectively _h, μ _srepresent the average saliency value of all pixels in the remarkable figure of two width respectively, the span of ρ is [-1,1], significantly scheme to be correlated with completely when ρ=1 represents two width, ρ=0 represents two width and significantly schemes completely uncorrelated, ρ=-1 represents two width figure inverse correlations, and the marking area namely in the remarkable figure of a width is completely not remarkable in the remarkable figure of another width.

Described step (4) comprises following concrete steps further:

(4.1) account for the image of the 1st weight for brightness, adopt the optimization histogram equalization method based on the remarkable figure of human eye;

(4.2) account for the image of the 1st weight for color characteristic, then take to strengthen the image enchancing method of color saturation in visual impression region-of-interest and contrast;

(4.3) account for the image of the 1st weight for textural characteristics, adopt the method for wavelet transformation here, denoising enhancing is carried out to visual impression region-of-interest, highlight the details of texture part.

The algorithmic formula adopted in described step (1.2) is as follows:

$f_c(x,y)=\frac{1}{(s_c+\exp (-\frac{saturation(x,y)}{{\mathrm{saturation}}_{a\mathrm{\ν}e}}\left)\right)\·(b_c+\exp (-\frac{brightness(x,y)}{{\mathrm{brightness}}_{a\mathrm{\ν}e}}\left)\right)}$

Wherein, saturation (x, y) and brightness (x, y) refers to saturation degree and the brightness value of pixel (x, y) respectively, saturation _aveand brightness _avethen represent the saturation degree of entire image and the mean value of brightness, s _cand b _cthat constant all gets 0.5.

The algorithmic formula that described step (1.3) adopts is as follows:

$O=\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2})\·cos(w\·x\·cos(\mathrm{\θ})+w\·y\·sin(\mathrm{\θ})+\frac{\mathrm{\π}}{2})$

Wherein, w is constant, and θ is direction θ=n π/k (n=0,1 of wave filter ... k-1), k refers to the direction number of Gabor filter; Extract the direction character of image with Gabor function, the direction setting of its wave filter is then 4 width characteristic patterns are formed final direction according to equal weight superposition significantly to scheme (O).

The algorithmic formula that described step (1.4) adopts is as follows:

$g(x,y)=\frac{k}{{\mathrm{\σ}}^2}\exp (-\frac{k^2(x^2+y^2)}{2{\mathrm{\σ}}^2})\exp \left(k\right(xcos\mathrm{\θ}+ysin\mathrm{\θ}\left)\right)$

Wherein, n refers to yardstick, and θ is finger direction.

The algorithmic formula that described step (2.4) adopts is as follows:

S＝w ₁C+w ₂I+w ₃O+w ₄T+w ₅E

In formula, w ₁, w ₂, w ₃, w ₄, w ₅be weighted value corresponding to each vision low-level image feature index respectively, add up to 1; C, I, O, T, E represent the remarkable figure of color, brightness, direction, texture and edge feature respectively; Finally namely S image is calculated remarkable figure to normalize to [0,255], and be processed into bianry image.

The invention has the beneficial effects as follows:

The present invention compares with conventional images Enhancement Method following advantage: the eye based on human eye fixation point is moved remarkable figure and is integrated into traditional image enchancing method by the present invention first, propose a kind of Itti visual attention model of view-based access control model low-level image feature of optimization to extract the calculating characteristic remarkable picture of image, improve visual quality when coloured image reproduces on digital distance scope, for different images type, highlight the important information in image, weaken secondary information, final image enhancement effects is made more to meet human eye visual perception, using value is played in image enhaucament.

Accompanying drawing explanation

Fig. 1 is the remarkable figure computation model overall flow figure of view-based access control model low-level image feature;

Fig. 2 is the final effect figure of two kinds of remarkable figure;

Fig. 3 is optimal weight analysis result table;

Fig. 4 is the histogram equalization schematic diagram of view-based access control model area-of-interest;

Fig. 5 is brightness image enchancing method comparison diagram;

Fig. 6 is that color and brightness substep carry out image enchancing method comparison diagram;

Fig. 7 is that texture and brightness substep carry out image enchancing method comparison diagram.

Specific implementation method

Do detailed elaboration below in conjunction with the preferred embodiment of accompanying drawing to the image enchancing method of a kind of view-based access control model low-level image feature involved in the present invention, but the present invention is not limited only to this embodiment.To have the present invention to make the public and understand thoroughly, in following preferred embodiment, carried out concrete details explanation.

1. the remarkable figure computation model of view-based access control model low-level image feature

As shown in Figure 1, the remarkable figure computation model overall flow of view-based access control model low-level image feature is as follows:

(1) brightness: be chosen in hsv color space and extract brightness, changes into the gray-scale map only having monochrome information by source images, finally image pixel value is normalized to again the scope of [0,255];

(2) color characteristic: color characteristic extracts under the hsv color space meeting human eye visual perception, after having extracted color characteristic, will normalize to the scope of [0,255], its algorithmic formula again by pixel value:

$f_c(x,y)=\frac{1}{(s_c+\exp (-\frac{saturation(x,y)}{{\mathrm{saturation}}_{a\mathrm{\ν}e}}\left)\right)\·(b_c+\exp (-\frac{brightness(x,y)}{{\mathrm{brightness}}_{a\mathrm{\ν}e}}\left)\right)}$

Wherein, saturation (x, y) and brightness (x, y) refers to saturation degree and the brightness value of pixel (x, y) respectively, saturation _aveand brightness _avethen represent the saturation degree of entire image and the mean value of brightness, s _cand b _cthat constant all gets 0.5.

(3) direction character: direction character is the local feature of image, showed some pixel in image also exist on certain direction in Rankine-Hugoniot relations, and then create a kind of visual sense of direction.General Bian Gabor transformation extracts the direction character of image, its computing formula:

$O=\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2})\·cos(w\·x\·\cos (\mathrm{\θ})+w\·y\·sin(\mathrm{\θ})+\frac{\mathrm{\π}}{2})$

Wherein, w is constant, and θ is direction θ=n π/k (n=0,1 of wave filter ... k-1), k refers to the direction number of Gabor filter.Adopt Gabor function to extract the direction character of image herein, the direction setting of its wave filter is then 4 width characteristic patterns are formed final direction according to equal weight superposition significantly to scheme (O).

(4) textural characteristics: adopt Gabor filter group to extract the feature of texture.First utilize Gabor filter group to extract textural characteristics on 5 yardsticks of image, 4 directions, then this 20 width textural characteristics figure is normalized, finally all images is superimposed by equal weight, forms last texture and significantly scheme; The algorithmic formula adopted is as follows:

$g(x,y)=\frac{k}{{\mathrm{\σ}}^2}\exp (-\frac{k^2(x^2+y^2)}{2{\mathrm{\σ}}^2})\exp \left(k\right(xcos\mathrm{\θ}+ysin\mathrm{\θ}\left)\right)$

Wherein, n refers to yardstick, and θ is finger direction.

(5) edge feature: the most typical extracting mode of edge feature is the Laplace operator utilizing second-order differential, extracts the gray-scale edges feature of image in HSV space.

After obtaining the characteristic pattern of 5 kinds of vision low-level image features, utilize the concrete steps of Itti model calculating characteristic remarkable picture as described below.

(1) Gaussian function is adopted respectively the color of input picture, brightness, direction, texture and edge to be carried out to the filtering of 6 yardsticks;

(2) Image Visual Feature under utilizing difference operation and Laplace operator to extract different scale;

(3) carry out central peripheral difference respectively to calculate, error image normalization is sued for peace, obtains the remarkable figure of this feature; (4) weight configuration (ensure that the weighted value of 5 vision low-level image features is added and equal 1) is carried out for the 5 width characteristic remarkable pictures obtained.Superposed by 5 width characteristic remarkable pictures, form the remarkable figure after 1 width combination of multiple features, formula is:

S＝w ₁C+w ₂I+w ₃O+w ₄T+w ₅E

In formula, w ₁, w ₂, w ₃, w ₄, w ₅be weighted value corresponding to each vision low-level image feature index respectively, add up to 1; C, I, O, T, E represent the remarkable figure of color, brightness, direction, texture and edge feature respectively; Finally namely S image is calculated remarkable figure to normalize to [0,255], and be processed into bianry image.

2. the remarkable figure computation model of the Gauss of human eye fixation point

Blinkpunkt generally refers to the position of human eye area-of-interest, the blinkpunkt obtaining human eye is caught by high-accuracy eye tracker, by Gaussian function to all blinkpunkt simulating two-dimensional Gaussian distribution, namely the value when blinkpunkt position is maximum, along with its field constantly expands, value will constantly reduce to be close to zero.Dimensional Gaussian distribution function is:

$H(x,y)=\frac{1}{2{\mathrm{\π\σ}}^2}\exp (-\frac{{(x-x_0)}^2+{(y-y_0)}^2}{2{\mathrm{\σ}}^2})$

In formula, H (x, y) is for blinkpunkt is to the influence degree of its neighboring area; (x ₀, y ₀) be the coordinate of blinkpunkt; σ refers to the variance parameter of gauss of distribution function.According to actual observation Distance geometry angle in experiment, parameter σ=37, all take into account all blinkpunkts.The final effect figure of two kinds of remarkable figure as shown in Figure 2, to be wherein former figure, Fig. 2 b significantly scheme S for calculating to Fig. 2 a, and to be that eye is dynamic significantly scheme H to Fig. 2 c.

3. based on the determination of different images vision low-level image feature weight coefficient

Adopt correlation analysis to compare and calculate the similarity significantly scheming to move with eye remarkable figure, thus determine further in the weight calculating each vision low-level image feature in remarkable figure.

The formula of related coefficient is as follows:

$\mathrm{\ρ}=\frac{{\mathrm{\Σ}}_x\left(H\right(x)-{\mathrm{\μ}}_H)\·\left(S\right(x)-{\mathrm{\μ}}_S)}{\sqrt{{\mathrm{\Σ}}_x{\left(H\right(x)-{\mathrm{\μ}}_H)}^2\·{\mathrm{\Σ}}_x{\left(S\right(x)-{\mathrm{\μ}}_S)}^2}}$

In formula, the pixel in x representative image, H (x), S (x) are that eye moves the saliency value significantly scheming and calculate current pixel in remarkable figure, μ respectively _h, μ _srepresent the average saliency value of all pixels in the remarkable figure of two width respectively.The span of ρ is [-1,1], significantly scheme to be correlated with completely when ρ=1 represents two width, ρ=0 represents two width and significantly schemes completely uncorrelated, ρ=-1 represents two width figure inverse correlations, and the marking area namely in the remarkable figure of a width is completely not remarkable in the remarkable figure of another width.

4. the result of calculation of optimal weight and analysis

It is objective and changeless that eye in experiment moves remarkable figure, and calculate remarkable figure and constantly can be adjusted by the weight allocation of visual signature, until a stack features weight makes to calculate significantly figure and the eye related coefficient moved between remarkable figure reach maximum, this group weighted value is exactly optimal weight.In order to reduce calculated amount, first by observing the hotspot graph that eye tracker experiment obtains, determine the several characteristic indexs the most deep on all types of picture impact, and correspondingly give larger weight, then the heavy accuracy value of weighting is 0.05, namely the weighted value scope of each visual signature is [0.05,0.75], and entitlement weight values adds up to 1.Obtain the possibility of the entitlement code reassignment met under these conditions, and calculate related coefficient, that group weighted value when selecting related coefficient maximum.Experimental result as shown in Figure 3.

As can be seen from Figure 3, when getting best weights weight values, the related coefficient between the remarkable figure of 2 width all reaches more than 80%, and correlativity is very strong.Mix from dividing of weighted value, for the image of personal portrait class, its textural characteristics weight reaches 0.46, and influence degree is maximum; And in natural land image, the weight of brightness reaches 0.42, to having the greatest impact of visual impression region-of-interest; In animal class image, the weight of textural characteristics reaches 0.34, and influence degree is maximum; In the image of color class, color characteristic has been endowed larger weight and has reached 0.34, and influence degree is maximum.

Comparing result shows, in dissimilar image, the influence degree of different visual signatures to interesting image regions is different; Same visual signature is also different to dissimilar image contributions degree.

5. based on the image enchancing method of different images vision low-level image feature

Drawn by Fig. 3, these 3 features of color, brightness and texture account for major weight in all types of image.Account for the image of the 1st weight for brightness, propose the optimization histogram equalization method based on the remarkable figure of human eye, principle is shown in Fig. 4.First constructing the gray level histogram based on the remarkable figure of human eye, then carrying out rational limited amplitude process to accounting for leading gray level in histogram; By the process saved space of deutomerite according to the quantity of information distribution situation of gray level histogram, optimize the gray level distributing to non-dominant.Like this, just reach the object of so-called proportion optimizing, experiment proves that the image enhancement effects of the method more meets human eye visual perception.

Account for the image of the 1st weight for color characteristic, then take to strengthen the image enchancing method of color saturation in visual impression region-of-interest and contrast.Account for the image of the 1st weight for textural characteristics, adopt the method for wavelet transformation here, denoising enhancing is carried out to visual impression region-of-interest, highlight the details of texture part.

Consider and there will be the close special circumstances of 2 principal visual low-level image feature weights, the weighted value as direction and brightness in the class image of ocean is very close.Thus setting (comprises 0.05) when the difference of 2 weights within 0.05, then carry out according to the corresponding Enhancement Method substep of the 1st, the 2nd weight.If the difference of 2 weights is more than 0.05, then still process according to the corresponding Enhancement Method of the 1st weight.

Fig. 5-7 contrasts the design sketch that other do not consider the image enchancing method of visual impression region-of-interest

As seen from Figure 5, when strengthening a kind of vision low-level image feature separately, the effect of the histogram enhancement method based on human eye visual perception of proposition is better, avoids the problem that traditional histogram equalization crosses enhancing.Wherein, Fig. 5 a is the gray-scale map of former Fig. 1 component, and Fig. 5 b is that traditional histogram equalization strengthens design sketch, and Fig. 5 c strengthens design sketch for optimizing histogram equalization.As can be seen from Figures 6 and 7, the effect of carrying out substep enhancing for multiple key feature proposed, only process for single features than traditional Enhancement Method, design sketch after final enhancing more can highlight the very important visual information (raising as all corresponding in color saturation, contrast in human eye area-of-interest, texture is more clear), meet human eye visual perception.Wherein, Fig. 6 a, Fig. 7 a are former figure, Fig. 6 b, Fig. 7 b is traditional images Enhancement Method, and Fig. 6 c, Fig. 7 c are the image enchancing method proposed.

Claims (9)

1. an image enchancing method for view-based access control model low-level image feature, is characterized in that, the method has following steps:

(1) extract the vision low-level image feature of image, comprise color, brightness, direction, texture and edge feature;

(2) Weighted Fusion becomes to calculate significantly figure;

(3) eye recorded with eye tracker moves remarkable figure to carry out related coefficient and compares, to determine the optimal weight of each vision low-level image feature;

(4) be that dissimilar image selects suitable image enchancing method according to the size of weight.

2. the image enchancing method of view-based access control model low-level image feature according to claim 1, is characterized in that, described step (1) comprises following concrete steps further:

(1.1) in hsv color space, extract brightness, source images is changed into the gray-scale map only having monochrome information, finally image pixel value is normalized to again the scope of [0,255];

(1.2) under the hsv color space meeting human eye visual perception, extract color characteristic, after having extracted color characteristic, pixel value is normalized to again the scope of [0,255];

(1.3) direction character of image is extracted by Gabor transformation;

(1.4) Gabor filter group texture feature extraction is adopted: first utilize Gabor filter group to extract textural characteristics on 5 yardsticks of image, 4 directions, then this 20 width textural characteristics figure is normalized, finally all images are superimposed by equal weight, form last texture and significantly scheme;

(1.5) the most typical extracting mode of edge feature is the Laplace operator utilizing second-order differential, extracts the gray-scale edges feature of image in HSV space.

3. the image enchancing method of view-based access control model low-level image feature according to claim 1, is characterized in that, described step (2) comprises following concrete steps further:

(2.1) Gaussian function is adopted respectively the color of input picture, brightness, direction, texture and edge to be carried out to the filtering of 6 yardsticks;

(2.2) Image Visual Feature under utilizing difference operation and Laplace operator to extract different scale;

(2.3) carry out central peripheral difference respectively to calculate, error image normalization is sued for peace, obtains the remarkable figure of this feature;

(2.4) carry out weight configuration for the 5 width characteristic remarkable pictures obtained, and ensure that the weighted value addition of 5 vision low-level image features equals 1; 5 width characteristic remarkable pictures are superposed, forms the remarkable figure after 1 width combination of multiple features.

4. the image enchancing method of view-based access control model low-level image feature according to claim 1, is characterized in that, the algorithmic formula that described step (3) adopts is as follows:

$\mathrm{\ρ}=\frac{{\mathrm{\Σ}}_x\left(H\right(x)-{\mathrm{\μ}}_H)\·\left(S\right(x)-{\mathrm{\μ}}_s)}{\sqrt{{\mathrm{\Σ}}_x{\left(H\right(x)-{\mathrm{\μ}}_H)}^2\·{\Σ}_x{\left(S\right(x)-{\mathrm{\μ}}_S)}^2}}$

In formula, the pixel in x representative image, H (x), S (x) are that eye moves the saliency value significantly scheming and calculate current pixel in remarkable figure, μ respectively _h, μ _srepresent the average saliency value of all pixels in the remarkable figure of two width respectively, the span of ρ is [-1,1], significantly scheme to be correlated with completely when ρ=1 represents two width, ρ=0 represents two width and significantly schemes completely uncorrelated, ρ=-1 represents two width figure inverse correlations, and the marking area namely in the remarkable figure of a width is completely not remarkable in the remarkable figure of another width.

5. the image enchancing method of view-based access control model low-level image feature according to claim 1, is characterized in that, described step (4) comprises following concrete steps further:

(4.1) account for the image of the 1st weight for brightness, adopt the optimization histogram equalization method based on the remarkable figure of human eye;

(4.2) account for the image of the 1st weight for color characteristic, then take to strengthen the image enchancing method of color saturation in visual impression region-of-interest and contrast;

(4.3) account for the image of the 1st weight for textural characteristics, adopt the method for wavelet transformation here, denoising enhancing is carried out to visual impression region-of-interest, highlight the details of texture part.

6. the image enchancing method of view-based access control model low-level image feature according to claim 2, is characterized in that, the algorithmic formula adopted in described step (1.2) is as follows:

$f_c(x,y)=\frac{1}{\left(s_c+\exp \left(-\frac{saturation(x,y)}{{\mathrm{saturation}}_{a\mathrm{\ν}e}}\right)\right)\·\left(b_c+\exp \left(-\frac{brightness(x,y)}{{\mathrm{brightness}}_{a\mathrm{\ν}e}}\right)\right)}$

Wherein, saturation (x, y) and brightness (x, y) refers to saturation degree and the brightness value of pixel (x, y) respectively, saturation _aveand brightness _avethen represent the saturation degree of entire image and the mean value of brightness, s _cand b _cthat constant all gets 0.5.

7. the image enchancing method of view-based access control model low-level image feature according to claim 2, is characterized in that, the algorithmic formula that described step (1.3) adopts is as follows:

$O=\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2})\·cos\left(w\·x\·cos\left(\mathrm{\θ}\right)+w\·y\·sin\left(\mathrm{\θ}\right)+\frac{\mathrm{\π}}{2}\right)$

Wherein, w is constant, and θ is direction θ=n π/k (n=0,1 of wave filter ... k-1), k refers to the direction number of Gabor filter; Extract the direction character of image with Gabor function, the direction setting of its wave filter is then 4 width characteristic patterns are formed final direction according to equal weight superposition significantly to scheme (O).

8. the image enchancing method of view-based access control model low-level image feature according to claim 2, is characterized in that, the algorithmic formula that described step (1.4) adopts is as follows:

$g(x,y)=\frac{k}{{\mathrm{\σ}}^2}\exp (-\frac{k^2(x^2+y^2)}{2{\mathrm{\σ}}^2})\exp \left(k\right(xcos\mathrm{\θ}+ysin\mathrm{\θ}\left)\right)$

Wherein, n refers to yardstick, and θ is finger direction.

9. the image enchancing method of view-based access control model low-level image feature according to claim 3, is characterized in that, the algorithmic formula that described step (2.4) adopts is as follows:

S＝w ₁C+w ₂I+w ₃O+w ₄T+w ₅E

In formula, w ₁, w ₂, w ₃, w ₄, w ₅be weighted value corresponding to each vision low-level image feature index respectively, add up to 1; C, I, O, T, E represent the remarkable figure of color, brightness, direction, texture and edge feature respectively; Finally namely S image is calculated remarkable figure to normalize to [0,255], and be processed into bianry image.