CN103106644B - Overcome the self-adaptation picture quality enhancement method of coloured image inhomogeneous illumination - Google Patents

Abstract

The invention discloses a kind of self-adaptation picture quality enhancement method overcoming coloured image inhomogeneous illumination.The method first calculates luminance picture; The average intensity value in luminance picture in neighborhood of pixels is utilized to calculate local luminance index; Set up the mapping relations between local luminance index and contrast strengthen coefficient; With contrast difference measurement between contrast strengthen coefficients to construct single channel image pixel, and then weighting is integrated into overall self-adaptation contrast energy term; By the cost minimization model of this overall adaptive energy and the image intensifying of entropy deviation item composition diagram; Gradient descent method is finally utilized to solve the minimum value of this model as the single channel image after enhancing; After all passages are all disposed, the single channel image after all strengthening is merged into the coloured image of output.The present invention, effectively promoting picture contrast and while removing colour cast, can eliminating the even effect of bright and excessively dark uneven illumination, keeps the details integrality of object in light and shade region.

Description

Overcome the self-adaptation picture quality enhancement method of coloured image inhomogeneous illumination

Technical field

The invention belongs to image enhancement technique field, particularly a kind of self-adaptation picture quality enhancement method overcoming coloured image inhomogeneous illumination.

Background technology

Human visual system has the characteristic of local auto-adaptive, when observation of nature scene, the mankind can regulate light-inletting quantity by shrinking amplification pupil, and carry out the dynamic range compression of intensity of illumination by retina and cerebral cortex, finally tell the object under different light.But imaging device cannot adapt to the photoenvironment of various complex condition as human visual system, presents the realistic colour of object in natural scene, this is unfavorable for that we analyze image, identify in daily life and scientific research.Therefore, simulating human vision system overcomes coloured image inhomogeneous illumination, strengthens image quality adaptively and has important actual application value.

That contribute as far back as simulating human vision system aspect is Land and McCann, and it is theoretical that they propose famous Retinex by a series of Color perception experiment.Theoretical based on Retinex, the people such as Tao Li propose the non-linear enhancing algorithm (LiTao of a kind of comprehensive neighbor dependency, K.VijayanAsari, Anintegratedneighborhooddependentapproachfornonlinearenh ancementofcolorimages, ProceedingsoftheIEEEComputerSocietyInternationalConferen ceonInformationTechnology:CodingandComputing (ITCC'04), vol.2, pp.138-139, 2004), successively brightness and contrast is strengthened, to reach compression of dynamic range, improve the object of dark area visual effect, but the global contrast enhancing in algorithm can make that bright areas is brighter and dark area is darker, image detail is caused to lose.In addition, the same with multi-Scale Retinex Algorithm with single scale Retinex algorithm, these image enchancing methods based on Retinex theory all depend on the convolution algorithm between Gaussian function and former figure, this can increase the complexity of algorithm, and can cause halation phenomenon in the light and shade region that brightness changes greatly that has a common boundary.Chinese patent [200810116385.2] has invented a kind of fast colourful image enchancing method based on Retinex theory, by convolution algorithm is reduced to mean operation, reduce algorithm complex, but mean operation still can cause halation phenomenon, and in the method, the image that the maximal value in coloured image three passages is formed is processed, the image information comprised in the dark area at each component channel Medium and low intensity pixel place can be lost.Chinese patent [201110316982.1] has invented the pseudo-foreign fiber colour-image reinforcing method of cotton under a kind of uneven illumination and system, after coloured image is transformed into HSI space from rgb space, by minimizing to correct about the Retinex Variation Model of I component and Gamma, I component is strengthened, again the image after enhancing is converted back rgb space, although the method Variation Model avoids convolution algorithm, but simulating human vision system in the method, is not had to do self-adaptive processing for inhomogeneous illumination.

Summary of the invention

The object of the present invention is to provide a kind of self-adaptation picture quality enhancement method overcoming coloured image inhomogeneous illumination, construct the cost minimization model of color image enhancement, and the bright-dark degree of image pixel region is judged by calculating local luminance index, make model can do self-adaptive processing according to the local luminance of image, thus the details of coloured image after enhancing in light and shade region can both well be kept.

The technical solution realizing the object of the invention is: a kind of self-adaptation picture quality enhancement method overcoming coloured image inhomogeneous illumination, calculates luminance picture, utilizes the average intensity value in luminance picture in neighborhood of pixels to calculate local luminance index a (x); Set up the mapping relations between local luminance index a (x) and contrast strengthen coefficient lambda (x), i.e. λ (x)=1/ (1+exp (5-10a (x))); With contrast difference measurement between contrast strengthen coefficients to construct single channel image pixel; Single channel image overall situation self-adaptation contrast energy term is integrated into contrast difference measurement weighting between pixel; By the cost minimization model of this overall adaptive energy and the image intensifying of entropy deviation item composition diagram; Single channel image after all strengthening, as the single channel image after strengthening, after all passages are all disposed, is merged into the coloured image of output by the minimum value finally utilizing gradient descent method to solve this model.

Compared with prior art, its remarkable advantage is in the present invention: the characteristic of human visual system and variational technique combine by (1), can strengthen image adaptively according to the bright-dark degree of image local area.(2) while effectively promoting image quality contrast, the color error ratio of coloured image can be removed.(3) there is better dynamic range of images regulating power.

Below in conjunction with accompanying drawing, the present invention is described in further detail.

Accompanying drawing explanation

Fig. 1 is the process flow diagram that the present invention overcomes the self-adaptation picture quality enhancement method of coloured image inhomogeneous illumination.

Fig. 2 is the mapping relations schematic diagram in the present invention between local luminance index a (x) and contrast strengthen coefficient lambda (x).

Fig. 3 (a) is test pattern " clothes3 " figure (size is 637 × 468) in the embodiment of the present invention, and Fig. 3 (b) is the histogram of three passages of Fig. 3 (a).

Fig. 4 (a) is test pattern " clothes4 " figure (size is 637 × 468) in the embodiment of the present invention, and Fig. 4 (b) is the histogram of three passages of Fig. 4 (a).

Fig. 5 (a) is the result images after strengthening " clothes3 " figure in the embodiment of the present invention, and Fig. 5 (b) is the histogram of three passages of Fig. 5 (a).

Fig. 6 (a) is the result images after strengthening " clothes4 " figure in the embodiment of the present invention, and Fig. 6 (b) is the histogram of three passages of Fig. 6 (a).

Embodiment

The present invention overcomes the self-adaptation picture quality enhancement method of coloured image inhomogeneous illumination using variational technique and human visual system's characteristic as the modeling means of model and theoretical foundation.Variational technique is method conventional in image procossing, problem to be processed is depended in the selection of its energy functional, the solution making functional reach minimum value is exactly the last result images of problem, although variational technique is easier to understand and solves image processing problem, how being applied to the self-adaptation picture quality enhancement overcoming coloured image inhomogeneous illumination is a technical barrier.In addition, because the perception of human visual system to object detail depends on the locus residing for object, have locality, therefore the enhancing of contrast also should have locality, and the inventive method structure meets the cost minimization model of the color image enhancement of human visual system's characteristic.To secondary single channel image I: Ω → [0,1] any in coloured image, make x=(x ₁, x ₂) and y=(y ₁, y ₂) represent the coordinate of any two pixels in I, I ₀represent original image.According to contrast strengthen locality and the vision adaptive of human visual system, colour correction energy functional should meet following form:

E _w(I)＝C _w(I)+D(I)，(1)

Wherein contrast energy term C _w(I) form is weighting function play a part localization contrast to calculate, c (I (x), I (y)) is basic contrast difference measurement.Containing basic contrast variable in c (I (x), I (y)) $t=\frac{min\left(I\right(x),I\left(y\right))}{max\left(I\right(x),I\left(y\right))}=\frac{I\left(x\right)+I\left(y\right)-|I\left(x\right)-I\left(y\right)|}{I\left(x\right)+I\left(y\right)+|I\left(x\right)-I\left(y\right)|}$ Regularization form adopt regularization form to be owing to containing signed magnitude arithmetic(al), not easily differentiate in t, therefore introduce regularization variable ε, use $A_{\∈}\left(x\right)=x\frac{\arctan (x/\∈)}{\arctan (1/\∈)}-\frac{\∈}{2\arctan (1/\∈)}\log (1+\frac{x^2}{{\∈}^2})$ Signed magnitude arithmetic(al) A (x) after expression regularization=| x|.Due to existing two kinds of basic contrast difference measurements with the contrast energy term formed with be applicable to the partially dark region of process and partially bright region respectively, consider that human visual system can adaptive environment brightness, we judge the bright-dark degree of image pixel region by calculating local luminance index a (x), and utilize this index Design contrast strengthen coefficient lambda (x)=1/ (1+exp (5-10a (x))), with contrast difference measurement between contrast strengthen coefficients to construct single channel image pixel $c_{\∈}^{\mathrm{\λ}}\left(I\right(x),I(y\left)\right)=\mathrm{\λ}\left(x\right)t_{\∈}+(1-\mathrm{\λ}(x\left)\right)log\left(t_{\∈}\right),$ And then weighting is integrated into overall self-adaptation contrast energy term make the inventive method can strengthen image quality adaptively according to the local luminance of image when processing image.In addition, the color that the vision adaptive of human visual system makes the mankind observe meets the original realistic colour of scene as far as possible, and the mankind can be made the intensity of illumination that senses to be regulated near an average intermediate value, and introducing form in the model of therefore the inventive method is deviation energy term for measuring I (x) and I ₀distance between (x) and between I (x) and average illumination 1/2, wherein $d_{I_0,1/2}\left(I\right(x\left)\right)=d_1\left(I\right(x),I_0(x\left)\right)+d_2\left(I\right(x),1/2)$ Be two metric functions and.By overall self-adaptation contrast energy term with entropy deviation item be coupled together set up image enhaucament minimize cost function, that is:

$I*=\underset{I}{\arg \min }\frac{1}{4}\underset{x\∈\mathrm{\Ω}}{\Σ}\underset{y\∈\mathrm{\Ω}}{\Σ}w(x,y)c_{\∈}^{\mathrm{\λ}}\left(I\right(x),I(y\left)\right)+D_{\mathrm{\α},\mathrm{\β}}^{\mathrm{\ϵ}}\left(I\right),---\left(2\right)$

Seek minimum value by gradient descent method, then we can obtain the optimal estimation I of image I ^*, finally three channel image after enhancing are merged into the coloured image of output.

The concrete steps realizing foregoing are:

Step 1: calculate luminance picture.By a coloured image to be reinforced the image pixel intensities of each passage all normalizes to [0,1] in interval, luminance picture according to NTSC (National Television System Committee) criterion calculation coloured image: L=0.2989 × R+0.587 × G+0.144 × B, wherein R, G, B are respectively coloured image the image of RGB three passages;

Step 2: calculate local luminance indication index.According to luminance picture L, calculate with each pixel x=(x ₁, x ₂) centered by, size is image pixel intensities mean value a (x) in the neighborhood of 5 × 5 pixels, wherein (x ₁, x ₂) represent the coordinate of any one pixel x in L, the mode that the pixel at image border place is symmetrical is filled to the pixel in neighborhood, a (x) is local luminance index, span is [0,1], when 0≤a (x)≤1/3, think that pixel x is positioned at the region of low-light level, as 1/3 < a (x) < 2/3, think that pixel x is positioned at the region of intermediate light, when 2/3≤a (x)≤1, think that pixel x is positioned at the region of high brightness;

Step 3: calculate contrast strengthen coefficient.Local luminance index a (x) is utilized to calculate mapping relations λ (x)=1/ (1+exp (5-10a (x))) between contrast strengthen coefficient lambda (x);

Step 4: contrast difference measurement between structure single channel image pixel.Right the image I of certain passage, with contrast strengthen coefficient lambda (x) and basic contrast variable t _εcontrast difference measurement between structure pixel, its computing formula is: $c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(y\left)\right)=\mathrm{\&lambda;}\left(x\right)t_{\&Element;}+(1-\mathrm{\&lambda;}(x\left)\right)log\left(t_{\&Element;}\right),$

Wherein: $t_{\&Element;}=\frac{I\left(x\right)+I\left(y\right)-A_{\&Element;}\left(I\right(x)-I\left(y\right))}{I\left(x\right)+I\left(y\right)+A_{\&Element;}\left(I\right(x)-I\left(y\right))},$

In formula $A_{\&Element;}\left(x\right)=x\frac{arctan(x/\&Element;)}{arctan(1/\&Element;)}-\frac{\&Element;}{2arctan(1/\&Element;)}log(1+\frac{x^2}{{\&Element;}^2}),$ ε is regularization parameter;

Step 5: structure single channel image overall situation Weighted adaptive contrast energy term.By being weighted integration to contrast difference measurement between pixel, obtain self-adaptation contrast energy term, its computing formula is:

$C_{w,\&Element;}^{\mathrm{\&lambda;}}\left(I\right)=\frac{1}{4}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}\underset{y\&Element;\mathrm{\&Omega;}}{\&Sigma;}w(x,y)c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(y\left)\right),$

Wherein: $w(x,y)={\left(\right||x-y|{|}_2\underset{y\&Element;\mathrm{\&Omega;}}{\&Sigma;}\left(1/\left|\right|x-y|{|}_2\right))}^{-1}$ It is weighting function;

Step 6: form single channel image enhancing and minimize Cost Model (function).By coupling images overall situation Weighted adaptive contrast energy with entropy measure of dispersion that sets up single channel image enhancing minimizes cost function,

$I^{*}=\underset{I}{\arg \min }\frac{1}{4}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}\underset{y\&Element;\mathrm{\&Omega;}}{\&Sigma;}w(x,y)c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(y\left)\right)+D_{\mathrm{\&alpha;},\mathrm{\&beta;}}^{\mathrm{\&epsiv;}}\left(I\right),$

Wherein entropy measure of dispersion is defined as:

$D_{\mathrm{\&alpha;},\mathrm{\&beta;}}^{\mathrm{\&epsiv;}}\left(I\right)=\mathrm{\&alpha;}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}(\frac{1}{2}\log \frac{1}{2I\left(x\right)}-(\frac{1}{2}-I\left(x\right)\left)\right)+\mathrm{\&beta;}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}\left(I_0\right(x)\log \frac{I_0\left(x\right)}{I\left(x\right)}-(I_0\left(x\right)-I\left(x\right)\left)\right)$

I in formula ₀be original single channel image, α, β > 0 is for weighing about the metric function item of I and 1/2 with about I and I ₀the parameter of metric function item;

Step 7: solve by channels minimize.Utilize gradient descent method to solve this and minimize cost function, by the minimum value I tried to achieve ^*as the single channel image after enhancing.Whether the passage judging when pre-treatment is last passage of image, if not last passage, then repeats the image that step 4 to step 7 processes next passage, if last passage, then carry out step 8;

Step 8: coloured image synthesizes.After all passages all process, the single channel image after all strengthening is merged into the coloured image of output.

Below in conjunction with drawings and Examples, implementation process of the present invention is carried out following detailed description.Adopt in this embodiment 2 width sizes in the image data base of SimonFraserUniversity (http://www.cs.sfu.ca/ ~ colour/data/colour_constancy_test_images/mondrian/index. html) be 637 × 468 image " clothes3 " figure and " clothes4 " figure test, data in this database are the coloured images to different objects shooting under different illumination conditions, have the features such as colour cast, low contrast, low key tone, details are not outstanding.

As shown in Figure 1, first input " clothes3 " figure (or " clothes4 " figure) to be reinforced, as shown in Fig. 3 (a) (or Fig. 4 (a)), be designated as then following steps are carried out:

Step 1: calculate luminance picture.By coloured image the image pixel intensities of each passage all normalizes in [0,1] interval, the luminance picture according to NTSC (National Television System Committee) criterion calculation coloured image:

L＝0.2989×R+0.587×G+0.144×B，

Wherein R, G, B are respectively coloured image the image of RGB three passages;

Step 2: calculate local luminance indication index.According to luminance picture L, calculate with each pixel x=(x ₁, x ₂) centered by, size is image pixel intensities mean value in the neighborhood of 5 × 5 pixels the mode that the pixel at image border place is symmetrical is filled to the pixel in neighborhood.A (x) is local luminance index, span is [0,1], when 0≤a (x)≤1/3, think that pixel x is positioned at the region of low-light level, as 1/3 < a (x) < 2/3, think that pixel x is positioned at the region of intermediate light, when 2/3≤a (x)≤1, think that pixel x is positioned at the region of high brightness;

Step 3: calculate contrast strengthen coefficient.Set up mapping relations λ (x)=1/ (1+exp (5-10a (x))) between local luminance index a (x) and contrast strengthen coefficient lambda (x), make when pixel x is positioned at the region of low-light level, the value of λ (x) is less, when pixel x is positioned at the region of high brightness, the value of λ (x) is larger.

As shown in Figure 2;

Right below in the image of three passages carry out in step 4 to step 7 process respectively:

Step 4: contrast difference measurement between structure single channel image pixel.Right the image I of certain passage, with contrast strengthen coefficient lambda (x) and basic contrast variable t _εthe contrast difference measurement of structure adaptive local brightness:

$c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(y\left)\right)=\mathrm{\&lambda;}\left(x\right)t_{\&Element;}+(1-\mathrm{\&lambda;}(x\left)\right)log\left(t_{\&Element;}\right),$

Wherein:

$t_{\&Element;}=\frac{I\left(x\right)+I\left(y\right)-A_{\&Element;}\left(I\right(x)-I\left(y\right))}{I\left(x\right)+I\left(y\right)+A_{\&Element;}\left(I\right(x)-I\left(y\right))},$

$A_{\&Element;}\left(x\right)=x\frac{arctan(x/\&Element;)}{arctan(1/\&Element;)}-\frac{\&Element;}{2arctan(1/\&Element;)}log(1+\frac{x^2}{{\&Element;}^2}),$ Regularization parameter ε=1/20;

Step 5: structure single channel image overall situation Weighted adaptive contrast energy term.By being weighted integration to contrast difference measurement between pixel, obtain self-adaptation contrast energy term, its computing formula is:

$C_{w,\&Element;}^{\mathrm{\&lambda;}}\left(I\right)=\frac{1}{4}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}\underset{y\&Element;\mathrm{\&Omega;}}{\&Sigma;}w(x,y)c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(y\left)\right),$

Wherein:

$w(x,y)={\left(\right||x-y|{|}_2\underset{y\&Element;\mathrm{\&Omega;}}{\&Sigma;}\left(1/\left|\right|x-y|{|}_2\right))}^{-1}$ It is weighting function.

Due to the regulating action of contrast strengthen coefficient lambda (x), in the picture low-light level region with closer to, in the picture high brightness region with closer to;

Step 6: form single channel image enhancing and minimize cost function.By coupling images overall situation Weighted adaptive contrast energy with entropy measure of dispersion that sets up single channel image enhancing minimizes cost function:

$I*=\underset{I}{\arg \min }\frac{1}{4}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}\underset{y\&Element;\mathrm{\&Omega;}}{\&Sigma;}w(x,y)c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(y\left)\right)+D_{\mathrm{\&alpha;},\mathrm{\&beta;}}^{\mathrm{\&epsiv;}}\left(I\right),$

Wherein entropy measure of dispersion is defined as:

$D_{\mathrm{\&alpha;},\mathrm{\&beta;}}^{\mathrm{\&epsiv;}}\left(I\right)=\mathrm{\&alpha;}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}(\frac{1}{2}\log \frac{1}{2I\left(x\right)}-(\frac{1}{2}-I\left(x\right)\left)\right)+\mathrm{\&beta;}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}\left(I_0\right(x)\log \frac{I_0\left(x\right)}{I\left(x\right)}-(I_0\left(x\right)-I\left(x\right)\left)\right)$

I in formula ₀be original single channel image, α, β > 0 is for weighing about the metric function item of I and 1/2 with about I and I ₀the parameter of metric function item, in this embodiment, be set to α=255/253, β=1;

Step 7: solve by channels minimize.Utilize gradient descent method to solve this and minimize cost function, by the minimum value I tried to achieve ^*as the single channel image after enhancing.Whether the passage judging when pre-treatment is last passage of image, if not last passage, then repeats the image that step 4 to step 7 processes next passage, if last passage, then carry out step 8;

Step 8: coloured image synthesizes.After three passages all process, the single channel image after all strengthening is merged into the coloured image of output, as shown in Fig. 5 (a) (or Fig. 6 (a)).

Below in conjunction with Fig. 3 to Fig. 6, further illustrate the present invention by the effect assessment of embodiment.

As shown in Fig. 3 (a) He Fig. 4 (a), original " clothes3 " figure and " clothes4 " figure contain some grain details, and wherein " clothes3 " image has serious blue colour cast, are difficult to see the original color of underpants' thing of publishing picture; " clothes4 " pattern colour tuningout is dark, and contrast is low.It can also be seen that from Fig. 3 (b) and Fig. 4 (b), the histogram of this two width figure tri-passages is all very uneven, and most of pixel all concentrates between the strength range of 0 to 50.The enhancing result of Fig. 5 (a) for obtaining after the inventive method process Fig. 3 (a), as can be seen from the figure, original serious blue colour cast eliminates, present the realistic colour that clothing is original, the detail sections such as the fold texture on clothing also reduce out well.The enhancing result of Fig. 6 (a) for obtaining after the inventive method process Fig. 4 (a), as can be seen from the figure, originally gloomy picture becomes bright-coloured bright, and the details such as the color fringe on clothing is also high-visible, fold texture also restore well.In addition as can be seen from the histogram of two width result images, three passages (Fig. 5 (b) and Fig. 6 (b)) also, the distribution range of image pixel intensities has been stretched, and histogram is more even than original.

Claims (7)

1. overcome a self-adaptation picture quality enhancement method for coloured image inhomogeneous illumination, it is characterized in that calculating luminance picture, utilize the average intensity value in luminance picture in neighborhood of pixels to calculate local luminance index a (x); Set up the mapping relations between local luminance index a (x) and contrast strengthen coefficient lambda (x), i.e. λ (x)=1/ (1+exp (5-10a (x))); With contrast difference measurement between contrast strengthen coefficients to construct single channel image pixel; Single channel image overall situation Weighted adaptive contrast energy term is integrated into contrast difference measurement weighting between pixel; The single channel image of this overall Weighted adaptive contrast energy term and the image intensifying of entropy measure of dispersion composition diagram is strengthened and minimizes Cost Model; Last by channels minimize method for solving, the single channel image after all strengthening, as the single channel image after strengthening, after all passages are all disposed, is merged into the coloured image of output by the minimum value namely utilizing gradient descent method to solve this model;

Contrast difference measurement between described structure single channel image pixel: right the image I of certain passage, parameter represent coloured image to be reinforced, with contrast strengthen coefficient lambda (x) and basic contrast variable t _εthe contrast difference measurement of structure adaptive local brightness: $c_E^{\mathrm{\&lambda;}}\left(I\right(x),I(y\left)\right)=\mathrm{\&lambda;}\left(x\right)t_{\&Element;}+(1-\mathrm{\&lambda;}(x\left)\right)log\left(t_{\&Element;}\right),$ Wherein:

$t_{\&Element;}=\frac{I\left(x\right)+I\left(y\right)-A_{\&Element;}\left(I\right(x)-I(y\left)\right)}{I\left(x\right)+I\left(y\right)+A_{\&Element;}\left(I\right(x)-I(y\left)\right)},$

$A_{\&Element;}\left(x\right)=x\frac{\arctan (x/\&Element;)}{\arctan (1/\&Element;)}-\frac{\&Element;}{2\arctan (1/\&Element;)}log(1+\frac{x^2}{{\&Element;}^2}),$ ε is regularization parameter.

2. the self-adaptation picture quality enhancement method overcoming coloured image inhomogeneous illumination according to claim 1, is characterized in that the method calculating luminance picture is: by a coloured image to be reinforced the image pixel intensities of each passage all normalizes in [0,1] interval, and the luminance picture according to NTSC criterion calculation coloured image: L=0.2989 × R+0.587 × G+0.144 × B, wherein R, G, B are respectively coloured image the image of RGB three passages.

3. the self-adaptation picture quality enhancement method overcoming coloured image inhomogeneous illumination according to claim 1, it is characterized in that calculating local luminance refers to that calibration method is: according to luminance picture L, calculate centered by each pixel x, size is image pixel intensities mean value a (x) in the neighborhood of 5 × 5 pixels wherein (x ₁, x ₂) represent the coordinate of any one pixel x in L, the mode that the pixel at image border place is symmetrical is filled to the pixel in neighborhood, a (x) is local luminance index, span is [0,1], when 0≤a (x)≤1/3, think that pixel x is positioned at the region of low-light level, as 1/3 < a (x) < 2/3, think that pixel x is positioned at the region of intermediate light, when 2/3≤a (x)≤1, think that pixel x is positioned at the region of high brightness.

4. the self-adaptation picture quality enhancement method overcoming coloured image inhomogeneous illumination according to claim 1, it is characterized in that the method calculating contrast strengthen coefficient is: according to local luminance index a (x), contrast strengthen coefficient lambda (x)=1/ (1+exp (5-10a (x))), make when pixel x is positioned at the region of low-light level, the value of λ (x) is less, when pixel x is positioned at the region of high brightness, the value of λ (x) is larger.

5. the self-adaptation picture quality enhancement method overcoming coloured image inhomogeneous illumination according to claim 1, is characterized in that structure single channel image overall situation Weighted adaptive contrast energy term: with contrast difference measurement between pixel $c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(y\left)\right)=\mathrm{\&lambda;}\left(x\right)t_{\&Element;}+(1-\mathrm{\&lambda;}(x\left)\right)log\left(t_{\&Element;}\right)$ Weighting is integrated into new contrast energy term $C_{w,\&Element;}^{\mathrm{\&lambda;}}\left(I\right)=\frac{1}{4}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}\underset{y\&Element;\mathrm{\&Omega;}}{\&Sigma;}w(x,y)c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(y\left)\right),$ Wherein $w(x,y)={\left(\right||x-y|{|}_2\underset{y\&Element;\mathrm{\&Omega;}}{\&Sigma;}(1/||x-y|{|}_2))}^{-1}$ It is weighting function.

6. the self-adaptation picture quality enhancement method overcoming coloured image inhomogeneous illumination according to claim 1, is characterized in that forming single channel image enhancing minimizes Cost Model: by coupling single channel image overall situation Weighted adaptive contrast energy term with entropy measure of dispersion set up single channel image enhancing and minimize Cost Model:

$I^{*}=\underset{I}{\arg min}\frac{1}{4}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}\underset{y\&Element;\mathrm{\&Omega;}}{\&Sigma;}w(x,y)c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(y\left)\right)+D_{\mathrm{\&alpha;},\mathrm{\&beta;}}^{\mathrm{\&epsiv;}}\left(I\right),$

Wherein entropy measure of dispersion is defined as:

$D_{\mathrm{\&alpha;},\mathrm{\&beta;}}^{\mathrm{\&epsiv;}}\left(I\right)=\mathrm{\&alpha;}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}(\frac{1}{2}log\frac{1}{2I\left(x\right)}-(\frac{1}{2}-I\left(x\right)\left)\right)+\mathrm{\&beta;}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}\left(I_0\right(x)log\frac{I_0\left(x\right)}{I\left(x\right)}-(I_0\left(x\right)-I\left(x\right)\left)\right),$

In formula, I0 is original single channel image, and α, β > 0 is for weighing about the metric function item of I and 1/2 with about I and I ₀the parameter of metric function item, for weighting function.

7. the self-adaptation picture quality enhancement method overcoming coloured image inhomogeneous illumination according to claim 1, it is characterized in that by channels minimize method for solving be: utilize gradient descent method solve single channel image strengthen minimize Cost Model, by the minimum value I tried to achieve ^*as the single channel image after enhancing, whether the passage judging when pre-treatment is last passage of image, if not last passage, then contrast difference measurement between repetitive construct single channel image pixel, structure single channel image overall situation Weighted adaptive contrast energy term, formed single channel image strengthen minimize Cost Model and by channels minimize solution procedure to process the image of next passage, if last passage, then carry out coloured image synthesis.