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

Abstract

The invention discloses an adaptive image quality enhancement method for overcoming non-uniform illumination of color images. The method first calculates the brightness image; uses the average intensity value in the pixel neighborhood of the brightness image to calculate the local brightness index; establishes the mapping relationship between the local brightness index and the contrast enhancement coefficient; uses the contrast enhancement coefficient to construct a single-channel image pixel contrast difference measure , and then weighted and integrated into a global adaptive contrast energy item; the global adaptive energy and entropy deviation item constitute the cost minimization model of image enhancement; finally, the gradient descent method is used to solve the minimum value of the model as the enhanced single-channel image ; After all channels are processed, merge all enhanced single-channel images into an output color image. While effectively improving image contrast and removing color cast, the invention can eliminate the uneven illumination effect of over-brightness and over-darkness, and maintain the integrity of details of objects in bright and dark areas.

Description

Adaptive image quality enhancement method for overcoming non-uniform illumination of color images

technical field

The invention belongs to the technical field of image enhancement, in particular to an adaptive image quality enhancement method for overcoming non-uniform illumination of color images.

Background technique

The human visual system has the characteristic of local adaptation. When observing natural scenes, humans can adjust the amount of incoming light by constricting and dilating the pupil, and compress the dynamic range of light intensity through the retina and cerebral cortex, and finally distinguish objects under different lighting conditions. However, imaging equipment cannot adapt to the lighting environment under various complex conditions like the human visual system, and present the true colors of objects in natural scenes, which is not conducive to our image analysis and recognition in daily life and scientific research. Therefore, simulating the human visual system to overcome the non-uniform illumination of color images and adaptively enhance image quality has important practical application value.

Land and McCann were the first to make contributions to simulating the human visual system. They proposed the famous Retinex theory through a series of color perception experiments. Based on the Retinex theory, Tao Li et al. proposed a comprehensive neighborhood dependent nonlinear enhancement algorithm (LiTao, K.Vijayan Asari, An integrated neighborhood dependent approach for nonlinear enhancement of color images, Proceeding of the IEEE Computer Society International Conference on Information Technology: Coding and Computing (ITCC'04), vol. 2, pp. 138-0439), , the brightness and contrast are enhanced successively to achieve the purpose of compressing the dynamic range and improving the visual effect of dark areas. However, the global contrast enhancement in the algorithm will make bright areas brighter and dark areas darker, resulting in loss of image details. In addition, like the single-scale Retinex algorithm and the multi-scale Retinex algorithm, these image enhancement methods based on the Retinex theory all rely on the convolution operation between the Gaussian function and the original image, which will increase the complexity of the algorithm, and will cause problems at the junction of light and shade. Areas with large changes in brightness cause the halo phenomenon. Chinese patent [200810116385.2] invented a fast color image enhancement method based on the Retinex theory. By simplifying the convolution operation into the mean value operation, the complexity of the algorithm is reduced. However, the mean value operation still causes halo phenomenon, and the method Processing the image formed by the maximum value of the three channels of the color image will lose the image information contained in the dark area where the low-intensity pixels in each component channel are located. Chinese patent [201110316982.1] invented a method and system for color image enhancement of cotton pseudo-heterosexual fibers under uneven illumination. After converting the color image from RGB space to HSI space, by minimizing the Retinex variational model and Gamma of the I component Correction enhances the I component, and then converts the enhanced image back to RGB space. Although this method uses a variational model to avoid convolution operations, this method does not simulate the adaptive processing of the human visual system for non-uniform illumination.

Contents of the invention

The purpose of the present invention is to provide an adaptive image quality enhancement method that overcomes the non-uniform illumination of color images, constructs a cost minimization model for color image enhancement, and judges the brightness and darkness of the area where the image pixels are located by calculating the local brightness index, so that The model can do adaptive processing according to the local brightness of the image, so that the details of the color image can be well preserved in the bright and dark areas after enhancement.

The technical solution for realizing the purpose of the present invention is: an adaptive image quality enhancement method for overcoming non-uniform illumination of color images, calculating the brightness image, and calculating the local brightness index a(x) by using the average intensity value in the neighborhood of pixels in the brightness image; Establish the mapping relationship between the local brightness index a(x) and the contrast enhancement coefficient λ(x), that is, λ(x)=1/(1+exp(5-10a(x))); use the contrast enhancement coefficient to construct a single Contrast difference measurement between channel image pixels; weighted integration of the contrast difference measurement between pixels into a single-channel image global adaptive contrast energy item; the cost minimization model of image enhancement is composed of the global adaptive energy and entropy deviation item; finally using the gradient The descent method solves the minimum value of the model as an enhanced single-channel image. After all channels are processed, all enhanced single-channel images are combined into an output color image.

Compared with the prior art, the present invention has the following remarkable advantages: (1) Combining the characteristics of the human visual system with the variational technology, the image can be adaptively enhanced according to the brightness and darkness of the local area of the image. (2) While effectively improving the contrast of the image quality, the color deviation of the color image can be removed. (3) It has better image dynamic range adjustment capability.

The present invention will be described in further detail below in conjunction with the accompanying drawings.

Description of drawings

Fig. 1 is a flowchart of an adaptive image quality enhancement method for overcoming non-uniform illumination of a color image according to the present invention.

Fig. 2 is a schematic diagram of the mapping relationship between the local brightness index a(x) and the contrast enhancement coefficient λ(x) in the present invention.

Figure 3(a) is the test image "clothes3" (size 637×468) in the embodiment of the present invention, and Figure 3(b) is the histogram of the three channels in Figure 3(a).

Figure 4(a) is a test image "clothes4" (size 637×468) in the embodiment of the present invention, and Figure 4(b) is a histogram of the three channels in Figure 4(a).

Fig. 5(a) is the result image after enhancing the "clothes3" image in the embodiment of the present invention, and Fig. 5(b) is the histogram of the three channels in Fig. 5(a).

Fig. 6(a) is the result image after enhancing the "clothes4" image in the embodiment of the present invention, and Fig. 6(b) is the histogram of the three channels in Fig. 6(a).

detailed description

The self-adaptive image quality enhancement method for overcoming non-uniform illumination of color images in the present invention uses variational technology and human visual system characteristics as modeling means and theoretical basis of the model. Variational technology is a commonly used method in image processing. The choice of its energy functional depends on the problem to be processed. The solution that makes the functional reach the minimum value is the final result image of the problem. Although variational technology is easier to understand and solve image processing However, how to apply it to the adaptive image quality enhancement that overcomes the non-uniform illumination of color images is a technical problem. In addition, since the human visual system's perception of object details depends on the spatial position of the object and is localized, the enhancement of contrast should also be localized. The method of the present invention constructs a color image enhancement that conforms to the characteristics of the human visual system with the least cost model. For any pair of single-channel images in color images I:Ω→[0,1], Let x=(x ₁ , x ₂ ) and y=(y ₁ , y ₂ ) represent the coordinates of any two pixels in I, and I ₀ represents the original image. According to the contrast enhancement locality and visual adaptability of the human visual system, the color correction energy functional should satisfy the following form:

E _w (I) = C _w (I) + D (I), (1)

where the contrast energy term C _w (I) has the form weight function Playing the role of localized contrast calculation, c(I(x),I(y)) is the basic contrast difference measure. c(I(x),I(y)) contains basic contrast variables $t=\frac{mi\mathrm{no}\left(I\right(x),I\left(\mathrm{the\; y}\right))}{max\left(I\right(x),I\left(\mathrm{the\; y}\right))}=\frac{I\left(x\right)+I\left(\mathrm{the\; y}\right)-|I\left(x\right)-I\left(\mathrm{the\; y}\right)|}{I\left(x\right)+I\left(\mathrm{the\; y}\right)+|I\left(x\right)-I\left(\mathrm{the\; y}\right)|}$ The regularized form of The regularization form is adopted because t contains absolute value operations, and it is not easy to obtain derivatives. Therefore, the regularization variable ε is introduced and used $A_{\∈}\left(x\right)=x\frac{\arctan (x/\∈)}{\arctan (1/\∈)}-\frac{\∈}{2\arctan (1/\∈)}\log (1+\frac{x^2}{{\∈}^2})$ Indicates the regularized absolute value operation A(x)=|x|. Since the existing two basic contrast difference measures and Contrast Energy Term and It is suitable for dealing with darker areas and brighter areas respectively. Considering that the human visual system can adapt to the brightness of the environment, we calculate the local brightness index a(x) to judge the brightness of the area where the image pixel is located, and use this index to design the contrast Enhancement coefficient λ(x)=1/(1+exp(5-10a(x))), using the contrast enhancement coefficient to construct a contrast difference measure between pixels of a single-channel image $c_{\∈}^{\mathrm{\λ}}\left(I\right(x),I(\mathrm{the\; y}\left)\right)=\mathrm{\λ}\left(x\right)t_{\∈}+(1-\mathrm{\λ}(x\left)\right)log\left(t_{\∈}\right),$ Then weighted and integrated into a global adaptive contrast energy item The method of the present invention can adaptively enhance the image quality according to the local brightness of the image when processing the image. In addition, the visual adaptability of the human visual system makes the colors observed by humans conform to the original real colors of the scene as much as possible, and enables humans to adjust the sensed light intensity around an average median value. Therefore, the model of the method of the present invention introduces in the form of The dispersion energy term of is used to measure the distance between I(x) and I ₀ (x) and between I(x) and the average illumination 1/2, where $d_{I_0,1/2}\left(I\right(x\left)\right)=d_1\left(I\right(x),I_0(x\left)\right)+d_2\left(I\right(x),1/2)$ is the sum of the two dispersion functions. The global adaptive contrast energy term and entropy deviation term Coupled to establish the minimum cost function of image enhancement, namely:

$\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; *</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; I</span>}}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; min</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{<span\; class="notranslate">\; \&Element;</span>}^{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}}^{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

By seeking the minimum value through the gradient descent method, we can obtain the optimal estimate I ^* of the image I, and finally merge the enhanced three-channel images into an output color image.

The specific steps to achieve the above content are:

Step 1: Compute the brightness image. A color image to be enhanced The pixel intensity of each channel is normalized to the [0,1] interval, and the brightness image of the color image is calculated according to the NTSC (National Television System Committee) standard: L=0.2989×R+0.587×G+0.144×B, where R , G, B are color images respectively The image of the three channels of red, green and blue;

Step 2: Calculate the local brightness indicator. According to the brightness image L, calculate the average pixel intensity a(x) in the neighborhood of each pixel point x=(x ₁ ,x ₂ ) with a size of 5×5 pixels, Where (x ₁ , x ₂ ) represents the coordinates of any pixel x in L, and the pixels at the edge of the image are filled with pixels in the neighborhood in a symmetrical manner, a(x) is the local brightness index, and the value range is [0,1], when 0≤a(x)≤1/3, the pixel point x is considered to be in the area of low brightness; when 1/3<a(x)<2/3, the pixel point x is considered to be in the In the area of medium brightness, when 2/3≤a(x)≤1, the pixel point x is considered to be in the area of high brightness;

Step 3: Calculate the contrast enhancement coefficient. Using the local brightness index a(x) to calculate the mapping relationship between the contrast enhancement coefficient λ(x) λ(x)=1/(1+exp(5-10a(x)));

Step 4: Construct a single-channel image pixel-to-pixel contrast difference metric. right For the image I of a certain channel, the contrast enhancement coefficient λ(x) and the basic contrast variable t _ε are used to construct the contrast difference measurement between pixels, and the calculation formula is: $c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(\mathrm{the\; y}\left)\right)=\mathrm{\&lambda;}\left(x\right)t_{\&Element;}+(1-\mathrm{\&lambda;}(x\left)\right)log\left(t_{\&Element;}\right),$

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

In the formula $A_{\&Element;}\left(x\right)=x\frac{arcta\mathrm{no}(x/\&Element;)}{arcta\mathrm{no}(1/\&Element;)}-\frac{\&Element;}{2arcta\mathrm{no}(1/\&Element;)}log(1+\frac{x^2}{{\&Element;}^2}),$ ε is the regularization parameter;

Step 5: Construct a globally weighted adaptive contrast energy term for a single-channel image. The adaptive contrast energy term is obtained by weighted integration of the inter-pixel contrast difference measure, and its calculation formula is:

${\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&Element;</span>}^{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{<span\; class="notranslate">\; \&Element;</span>}^{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

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

Step 6: Form a single-channel image enhancement minimization cost model (function). Adaptive contrast energy by coupling image global weighting and the entropy dispersion measure Establish a minimized cost function for single-channel image enhancement,

${\mathrm{<span\; class="notranslate">\; I</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; I</span>}}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; min</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{<span\; class="notranslate">\; \&Element;</span>}^{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}}^{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

where the entropy dispersion measure is defined as:

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}}^{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; log</span>}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; log</span>}\frac{{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>$

In the formula, I ₀ is the original single-channel image, and α, β > 0 are parameters used to weigh the dispersion function item about I and 1/2 and the dispersion function item about I and I ₀ ;

Step 7: Minimize solution channel by channel. The gradient descent method is used to solve the minimized cost function, and the obtained minimum value I ^* is used as the enhanced single-channel image. Determine whether the currently processed channel is the last channel of the image, if not the last channel, then repeat steps 4 to 7 to process the image of the next channel, if it is the last channel, go to step 8;

Step 8: Color image synthesis. After all channels are processed, all enhanced single-channel images are merged into an output color image.

The implementation process of the present invention will be described in detail below in conjunction with the accompanying drawings and embodiments. In this embodiment, two images with a size of 637×468 “clothes3” in the image database (http://www.cs.sfu.ca/～colour/data/colour_constancy_test_images/mondrian/index.html) of Simon Fraser University are used. " and "clothes4" images are used for experiments. The data in this database are color images taken of different objects under different lighting conditions, which have the characteristics of color cast, low contrast, low tone, and inconspicuous details.

As shown in Figure 1, first input the "clothes3" image (or "clothes4" image) to be enhanced, as shown in Figure 3(a) (or Figure 4(a)), and record it as Then proceed to the following steps:

Step 1: Compute the brightness image. color image The pixel intensity of each channel is normalized to the [0,1] interval, and the brightness image of the color image is calculated according to the NTSC (National Television System Committee) standard:

L=0.2989×R+0.587×G+0.144×B,

Where R, G, and B are color images respectively The image of the three channels of red, green and blue;

Step 2: Calculate the local brightness indicator. According to the brightness image L, calculate the average pixel intensity in the neighborhood of each pixel x=(x ₁ ,x ₂ ) with a size of 5×5 pixels For the pixels at the edge of the image, the pixels in the neighborhood are filled in a symmetrical manner. a(x) is the local brightness index, and the value range is [0,1]. When 0≤a(x)≤1/3, the pixel point x is considered to be in a low-brightness area, and when 1/3<a (x)<2/3, the pixel point x is considered to be in the area of medium brightness, and when 2/3≤a(x)≤1, the pixel point x is considered to be in the area of high brightness;

Step 3: Calculate the contrast enhancement coefficient. Establish a mapping relationship between the local brightness index a(x) and the contrast enhancement coefficient λ(x) λ(x)=1/(1+exp(5-10a(x))), so that when the pixel point x is located at low brightness When the pixel point x is located in a high brightness area, the value of λ(x) is small, and the value of λ(x) is large.

as shown in picture 2;

next to The images of the three channels in are processed in step 4 to step 7 respectively:

Step 4: Construct a single-channel image pixel-to-pixel contrast difference metric. right For an image I of a certain channel of , use the contrast enhancement coefficient λ(x) and the basic contrast variable t _ε to construct a contrast difference metric for adaptive local brightness:

${\mathrm{<span\; class="notranslate">\; c</span>}}_{<span\; class="notranslate">\; \&Element;</span>}^{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{<span\; class="notranslate">\; \&Element;</span>}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{<span\; class="notranslate">\; \&Element;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

in:

${\mathrm{<span\; class="notranslate">\; t</span>}}_{<span\; class="notranslate">\; \&Element;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{<span\; class="notranslate">\; \&Element;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{<span\; class="notranslate">\; \&Element;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>$

$A_{\&Element;}\left(x\right)=x\frac{arcta\mathrm{no}(x/\&Element;)}{arcta\mathrm{no}(1/\&Element;)}-\frac{\&Element;}{2arcta\mathrm{no}(1/\&Element;)}log(1+\frac{x^2}{{\&Element;}^2}),$ Regularization parameter ε=1/20;

Step 5: Construct a global weighted adaptive contrast energy term for a single-channel image. The adaptive contrast energy term is obtained by weighted integration of the contrast difference measures between pixels, and its calculation formula is:

${\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&Element;</span>}^{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{<span\; class="notranslate">\; \&Element;</span>}^{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

in:

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

Due to the adjustment effect of the contrast enhancement coefficient λ(x), In areas of low brightness in the image and Closer, the areas of high brightness in the image are associated with Closer;

Step 6: Form a single-channel image enhancement to minimize the cost function. Adaptive contrast energy by coupling image global weighting and the entropy dispersion measure Create a minimized cost function for single-channel image enhancement:

$\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; *</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; I</span>}}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; min</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{<span\; class="notranslate">\; \&Element;</span>}^{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}}^{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

where the entropy dispersion measure is defined as:

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}}^{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; log</span>}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; log</span>}\frac{{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>$

In the formula, I ₀ is the original single-channel image, α, β>0 is used to weigh the parameters of the dispersion function item about I and 1/2 and about the dispersion function item of I and I ₀ , in this embodiment Set to α=255/253, β=1;

Step 7: Minimize solution channel by channel. The gradient descent method is used to solve the minimized cost function, and the obtained minimum value I ^* is used as the enhanced single-channel image. Determine whether the currently processed channel is the last channel of the image, if not the last channel, then repeat steps 4 to 7 to process the image of the next channel, if it is the last channel, then proceed to step 8;

Step 8: Color image synthesis. After all three channels are processed, all enhanced single-channel images are merged into an output color image, as shown in Fig. 5(a) (or Fig. 6(a)).

The present invention will be further described through the effect evaluation of the embodiments below in conjunction with FIG. 3 to FIG. 6 .

As shown in Figure 3(a) and Figure 4(a), the original "clothes3" image and "clothes4" image both contain some texture details, and the "clothes3" image has a serious blue color cast, and it is difficult to see the clothes in the image Original colors; "clothes4" image has darker tones and low contrast. It can also be seen from Fig. 3(b) and Fig. 4(b) that the histograms of the three channels of these two images are very uneven, and most pixels are concentrated between the intensity range of 0 to 50. Fig. 5(a) is the enhancement result obtained after processing Fig. 3(a) by the method of the present invention. It can be seen from the figure that the original serious blue color cast has been removed, showing the original true color of the clothes, and the Details such as wrinkled textures are also well restored. Figure 6(a) is the enhancement result obtained after processing Figure 4(a) by the method of the present invention. It can be seen from the figure that the original gray and dark picture becomes bright and bright, and the colored stripes on the clothes are also clearly visible, and details such as wrinkle texture Also restored nicely. In addition, it can also be seen from the histograms of the three channels of the two result images (Figure 5(b) and Figure 6(b)), the distribution range of pixel intensity is stretched, and the histogram is more uniform than before.

Claims (7)

1. An adaptive image quality enhancement method for overcoming the non-uniform illumination of color images, characterized in that it calculates the brightness image, utilizes the average intensity value in the pixel neighborhood in the brightness image to calculate the local brightness index a(x); establishes the local brightness index a The mapping relationship between (x) and the contrast enhancement coefficient λ(x), that is, λ(x)=1/(1+exp(5-10a(x))); use the contrast enhancement coefficient to construct the contrast between pixels of a single-channel image Difference measurement; weighted integration of the contrast difference measurement between pixels into a single-channel image global weighted adaptive contrast energy item; the global weighted adaptive contrast energy item and the entropy deviation measurement form a single-channel image enhancement minimization cost model for image enhancement; Finally, the channel-by-channel minimization solution method is to use the gradient descent method to solve the minimum value of the model as the enhanced single-channel image. After all channels are processed, all the enhanced single-channel images are combined into an output color image; The contrast difference metric between the constructed single-channel image pixels: The image I of a certain channel of the parameter Represents the color image to be enhanced, using the contrast enhancement coefficient λ(x) and the basic contrast variable t _ε to construct a contrast difference metric for adaptive local brightness: $c_{\mathrm{E.}}^{\mathrm{\&lambda;}}\left(I\right(x),I(\mathrm{the\; y}\left)\right)=\mathrm{\&lambda;}\left(x\right)t_{\&Element;}+(1-\mathrm{\&lambda;}(x\left)\right)log\left(t_{\&Element;}\right),$ in: ${\mathrm{<span\; class="notranslate">\; t</span>}}_{<span\; class="notranslate">\; \&Element;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{<span\; class="notranslate">\; \&Element;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{<span\; class="notranslate">\; \&Element;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>$ $A_{\&Element;}\left(x\right)=x\frac{\arctan (x/\&Element;)}{\arctan (1/\&Element;)}-\frac{\&Element;}{2\arctan (1/\&Element;)}log(1+\frac{x^2}{{\&Element;}^2}),$ ε is a regularization parameter. 2. The adaptive image quality enhancement method for overcoming non-uniform illumination of color images according to claim 1, characterized in that the method for calculating the brightness image is: a color image to be enhanced The pixel intensity of each channel is normalized to the [0,1] interval, and the brightness image of the color image is calculated according to the NTSC standard: L=0.2989×R+0.587×G+0.144×B, where R, G, and B are respectively color image An image with three channels of red, green and blue. 3. The adaptive image quality enhancement method for overcoming non-uniform illumination of color images according to claim 1, characterized in that the method for calculating the local brightness index is: according to the brightness image L, calculate the center of each pixel point x, the size is the average pixel intensity a(x) in a neighborhood of 5×5 pixels, Where (x ₁ , x ₂ ) represents the coordinates of any pixel x in L, for the pixels at the edge of the image, the pixels in the neighborhood are filled in a symmetrical manner, a(x) is the local brightness index, and the value range is [0,1], when 0≤a(x)≤1/3, the pixel point x is considered to be in the area of low brightness; when 1/3<a(x)<2/3, the pixel point x is considered to be in the In the area of medium brightness, when 2/3≤a(x)≤1, the pixel point x is considered to be in the area of high brightness. 4. The adaptive image quality enhancement method for overcoming non-uniform illumination of color images according to claim 1, characterized in that the method for calculating the contrast enhancement coefficient is: according to the local brightness index a (x), the contrast enhancement coefficient λ (x) =1/(1+exp(5-10a(x))), so that when the pixel point x is located in a low-brightness area, the value of λ(x) is small, and when the pixel point x is located in a high-brightness area, λ The value of (x) is larger. 5. The adaptive image quality enhancement method for overcoming non-uniform illumination of color images according to claim 1, characterized in that it constructs a single-channel image global weighted adaptive contrast energy item: measured by contrast difference between pixels $c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(\mathrm{the\; y}\left)\right)=\mathrm{\&lambda;}\left(x\right)t_{\&Element;}+(1-\mathrm{\&lambda;}(x\left)\right)log\left(t_{\&Element;}\right)$ Weighted integration into a new contrast energy term $C_{w,\&Element;}^{\mathrm{\&lambda;}}\left(I\right)=\frac{1}{4}\underset{x\&Element;\mathrm{\&Omega;}}{\&Sigma;}\underset{\mathrm{the\; y}\&Element;\mathrm{\&Omega;}}{\&Sigma;}w(x,\mathrm{the\; y})c_{\&Element;}^{\mathrm{\&lambda;}}\left(I\right(x),I(\mathrm{the\; y}\left)\right),$ in $w(x,\mathrm{the\; y})={\left(\right||x-\mathrm{the\; y}|{|}_2\underset{\mathrm{the\; y}\&Element;\mathrm{\&Omega;}}{\&Sigma;}(1/||x-\mathrm{the\; y}|{|}_2))}^{-1}$ is the weight function. 6. The adaptive image quality enhancement method for overcoming non-uniform illumination of color images according to claim 1, characterized in that a single-channel image enhancement minimization cost model is formed: by coupling the single-channel image global weighted adaptive contrast energy term and the entropy dispersion measure Establish a single-channel image enhancement minimization cost model: ${\mathrm{<span\; class="notranslate">\; I</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; I</span>}}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{<span\; class="notranslate">\; \&Element;</span>}^{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}}^{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$ where the entropy dispersion measure is defined as: ${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}}^{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; g</span>}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; g</span>}\frac{{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$ In the formula, I0 is the original single-channel image, α, β>0 is used to weigh the parameters of the dispersion function item about I and 1/2 and the dispersion function item about I and I ₀ , is the weight function. 7. The adaptive image quality enhancement method for overcoming non-uniform illumination of color images according to claim 1, characterized in that the channel-by-channel minimization solution method is: using the gradient descent method to solve the single-channel image enhancement minimization cost model, and will obtain The obtained minimum value I ^* is used as the enhanced single-channel image to determine whether the currently processed channel is the last channel of the image. If it is not the last channel, repeat the construction of the contrast difference measurement between the pixels of the single-channel image and the global construction of the single-channel image. Weighting the adaptive contrast energy term, forming a single-channel image enhancement minimization cost model and channel-by-channel minimization solution step to process the image of the next channel, and if it is the last channel, color image synthesis.