CN104282004B - Self-adaptation equalization method based on extensible segmentation histogram - Google Patents

Abstract

The invention discloses an adaptive histogram equalization method based on scalable segmentation, which comprises the following steps: Step 1: using the scalable histogram segmentation method to expand the original input histogram; Step 2: expanding the expanded histogram The histogram is equalized, and each equalized sub-histogram has the same length as the original histogram; Step 3: The equalized sub-histogram obtained in step 2 is weighted and mixed according to the following formula to obtain the final equalization Histogram H ^t (i); Using this method, the scalable histogram segmentation strategy can be used to realize the gray redistribution of the equalized image to a larger dynamic range, which is conducive to the full enhancement of the image in the dark area; the application proposed The self-adaptive histogram equalization method, which adaptively controls the uniform distribution of the gray level according to the image gray level attribute, thus avoiding the over-enhancement, under-enhancement and unnatural halo phenomenon of the image; the calculation speed of the whole method is fast, and the obtained Enhanced images are noticeable.

Description

An Adaptive Equalization Method Based on Scalable Segmented Histogram

technical field

The invention relates to an adaptive equalization method based on scalable segmented histograms.

Background technique

Histogram equalization (HE) has been widely used in traffic monitoring, medical image processing and other fields due to its simple principle and fast calculation speed. HE achieves contrast enhancement by distributing high dynamic range gray levels to a uniform distribution to increase the difference between adjacent gray levels. However, if there are multiple peaks in the histogram, over-enhancement, under-enhancement, and unnatural halos will occur; in order to avoid the above problems, adaptive local histogram equalization (adaptive local histogram equalization, ALHE), contrast Contrast-limited adaptive histogram equalization (CLAHE) and other methods first divide the image into multiple small windows, and perform histogram equalization in the small windows respectively; but due to the lack of global information, these methods are often excessive or checkerboard lattice phenomenon. The histogram modification framework (HMF) regards image contrast enhancement as an optimization problem consisting of histogram uniform redistribution, original histogram preservation and histogram smooth combination, so as to achieve adjustable image contrast enhancement; two-dimensional Histogram equalization (two-dimensional histogram equalization, 2DHE) simultaneously considers the gray level distribution of each pixel and neighboring pixels to construct a two-dimensional histogram, and then uniformly redistributes the two-dimensional histogram to achieve image contrast enhancement; AMHE directly modifies the histogram of the original image, and uses the histogram specification method to achieve image enhancement; however, these methods are all for the entire histogram, and it is difficult to avoid over-enhancement and under-enhancement, and in flat areas of the image, there will be Halo phenomenon occurs. To this end, researchers have proposed many improved methods, such as brightness preserving histogram equalization (BBHE), same area sub-image histogram equalization (DSIHE), minimum average brightness error double histogram equalization (MMBEBHE), recursive mean separation histogram Equalization (RMSHE) and Brightness Preserving Motion Blur Histogram Equalization (BPDFHE), etc. In addition, a Gaussian mixture model is introduced to fit the gray distribution, and then the intersection points between the mixture models are used to separate the original histogram, and the histogram equalization is performed separately. These multi-histogram equalization methods can effectively maintain brightness, but when used for dark image enhancement, it will lead to under-enhancement; in addition, since multiple sub-histograms use the same equalization strategy, when the image has a large contrast between light and dark , these methods can easily lead to over-enhancement.

Contents of the invention

The present invention provides an adaptive equalization method based on scalable segmented histograms, and its purpose is to solve the problem of under-enhancement in dark areas of images and over-enhancement or severe seriousness in bright areas in the process of enhancing dark images in existing methods. degradation phenomena and other issues.

A method based on scalable segment adaptive histogram equalization, comprising the following steps:

Step 1: Extend the original input histogram using the scalable histogram segmentation method;

First, the original histogram H ^o is divided into N non-overlapping segmented histograms h _k , k∈[1,...,N]; then all the segmented histograms h _k are respectively extended to be as gray as the original histogram H ^o For the histogram with the same degree interval, keep the position of the segmented histogram h _k in the original histogram unchanged, and fill the other positions with zeros, that is, an expandable segmented histogram (k∈[1,...,N]), $h_k^x=[0,...,h_k,...,0],(k\∈[1,...,N\left]\right);$

Step 2: Equalize the scalable segmented histogram, and each equalized sub-histogram has the same length as the original histogram;

Step 3: The equalized sub-histogram obtained in step 2 Perform weighted mixing according to the following formula to obtain the final equalized histogram H ^t (i):

${\mathrm{<span\; class="notranslate">\; h</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>$

where w _k (i) is the normalized weight of the k-th equalized extended segment histogram, weight function N is the number of histogram segments, i is the gray level, u _k and σ _k are the mean and variance of the weight function respectively.

This method solves the problem of over-enhancement in bright areas and under-enhancement in dark areas.

The mean of the weight function in, are the first and last gray levels of the segmented histogram h _k respectively;

The variance of the weight function To avoid over-enhancing the image threshold, is the coverage width of the segmented histogram h _k , namely

It is guaranteed that the histogram h _k has the same distribution degree in the two symmetrical directions during the equalization process.

said for in, is the gray mean value in the segmented histogram h _k .

The equalization process selected in the step 2 is an adaptive equalization process, and the optimization problem proposed in the existing histogram equalization method In , the regularization parameters α, β and γ are optimized using the following formulas;

$\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; th</span>}}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; \&beta;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; th</span>}}<span\; class="notranslate">\; )</span>$

γ=1-α-β

Among them, H ^x is the expanded histogram of the input image, H ^t is the target histogram of the input image, ^Hu is the expected equalization distribution histogram; H is the histogram obtained after adaptive equalization processing, ||·|| Indicates the Euclidean norm, D is the double diagonal difference matrix;

Measure the distance between the gray level i of the expanded histogram H ^x of the input image and the maximum gray level I _max of the expanded histogram H ^x of the input image, i is the gray level in the expanded histogram H ^x of the input image Level, L is the number of gray levels, n _i is the frequency of gray level i in the expanded histogram H ^x of the input image, I _max is the maximum gray level, and the variance σ is 0.2I _max ;

m _i is the number of pixels whose gray value I(x _j ) of pixel x _j in the input image is equal to i; nG(x _j ) is the normalized local variance of pixel x _j , and nG(x _j )=G(x _j )/I(x _j ), G(x _j ) is the gradient magnitude at pixel x _j , G represents the gradient magnitude map corresponding to the input image I, where, Indicates the convolution operation, and S is the Sobel operator template.

The expected equilibrium distribution histogram H ^u is distributed using the following formula:

${\mathrm{<span\; class="notranslate">\; h</span>}}^{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>& \mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&NotElement;</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\end{array}\right.$

In the formula, i is the gray level.

After obtaining the final equalized histogram, the final enhanced image can be easily obtained through histogram correction.

Beneficial effect

The present invention proposes an adaptive histogram equalization method based on scalable segmentation, including the following steps: Step 1: expand the original histogram input by using the scalable histogram segmentation method; Step 2: expand the expanded histogram The histogram is equalized, and each equalized sub-histogram has the same length as the original histogram; Step 3: The equalized sub-histogram obtained in step 2 is weighted and mixed according to the following formula to obtain the final equalization Histogram H ^t (i); Using this method, the scalable histogram segmentation strategy can be used to realize the gray redistribution of the equalized image to a larger dynamic range, which is conducive to the full enhancement of the image in the dark area; the application proposed The self-adaptive histogram equalization method, which adaptively controls the uniform distribution of the gray level according to the image gray level attribute, thus avoiding the over-enhancement, under-enhancement and unnatural halo phenomenon of the image; the calculation speed of the whole method is fast, and the obtained Enhanced images are noticeable.

Description of drawings

Fig. 1 is the schematic diagram of the process of applying the method of the present invention to process image, wherein, figure (a) is the original image of input, and figure (b) is that original image is converted into HSV space image, and figure (c) is to figure (b) The V channel image in the figure is the histogram obtained by using the scalable histogram segmentation method. Figure (d) is the histogram obtained by adaptive equalization processing on the image (c), and the image (e) is the histogram obtained by the image (d). The histogram obtained by the weighting process, figure (f) is the enhanced image of the V channel obtained by utilizing the histogram of figure (e), and figure (g) is the final enhanced effect figure obtained;

Fig. 2 is the comparison diagram of applying the method of the present invention and other histogram equalization methods to enhance the same image, wherein, Fig. (a) is the original image, Fig. (b) adopts RMSHE method, and Fig. (c) is Adopt BPBFHE method, figure (d) is to adopt HMF method, figure (e) is to adopt 2DHE method, figure (f) is to adopt method AEPHE of the present invention, figure (g) is each histogram equalization method input and output interval Grayscale mapping function, the abscissa is the input grayscale, and the ordinate is the output grayscale;

FIG. 3 is a schematic diagram of the weight function W ₁ corresponding to the segmented histogram h _k when N=3.

Detailed ways

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

A method based on scalable segment adaptive histogram equalization, comprising the following steps:

Step 1: Extend the original input histogram using the scalable histogram segmentation method;

First, the original histogram H ^o is divided into N non-overlapping segmented histograms h _k , k∈[1,N]; then all segmented histograms h _k are respectively extended to the gray interval of the original histogram H ^o For the same histogram, keep the position of the segmented histogram h _k in the original histogram unchanged, and fill other positions with zeros, that is, get the extended segmented histogram (k∈[1,N]), $h_k^x=[0,...,h_k,...,0],(k\&Element;[1,...,N\left]\right);$

Step 2: Equalize the extended segmented histogram, and each equalized sub-histogram has the same length as the original histogram;

Optimization problem posed in existing histogram equalization methods In , the regularization parameters α, β and γ are optimized using the following formulas;

$\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; th</span>}}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; \&beta;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; th</span>}}<span\; class="notranslate">\; )</span>$

γ=1-α-β

Among them, H ^x is the expanded histogram of the input image, H ^t is the target histogram of the input image, ^Hu is the expected equalization distribution histogram; H is the histogram obtained after adaptive equalization processing, ||·|| Indicates the Euclidean norm, D is the double diagonal difference matrix;

Measure the distance between the gray level i of the expanded histogram H ^x of the input image and the maximum gray level I _max of the expanded histogram H ^x of the input image, i is the gray level in the expanded histogram H ^x of the input image Level, L is the number of gray levels, n _i is the frequency of gray level i in the expanded histogram H ^x of the input image, I _max is the maximum gray level, and the variance σ is 0.2I _max ;

m _i is the number of pixels whose gray value I(x _j ) of pixel x _j in the input image is equal to i; nG(x _j ) is the normalized local variance of pixel x _j , and nG(x _j )=G(x _j )/I(x _j ), G(x _j ) is the gradient magnitude at pixel x _j , in, Indicates the convolution operation, and S is the Sobel operator template.

The expected equilibrium distribution histogram H ^u is distributed using the following formula:

${\mathrm{<span\; class="notranslate">\; h</span>}}^{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>& \mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&NotElement;</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\end{array}\right.$

In the formula, i is the gray level.

After obtaining the final equalized histogram, the final enhanced image can be easily obtained through histogram correction.

Step 3: The equalized sub-histogram obtained in step 2 is weighted and mixed according to the following formula to obtain the final equalized histogram H ^t (i):

${\mathrm{<span\; class="notranslate">\; h</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>$

where w _k (i) is the normalized weight of the k-th equalized extended segment histogram, weight function k=[1,...,N], N is the number of histogram segments, i is the gray level, u _k and σ _k are the mean and variance of the weight function respectively.

The mean of the weight function in, are the first and last gray levels of the segmented histogram h _k respectively;

The variance of the weight function To avoid image over-enhancement threshold, is the coverage width of the segmented histogram h _k , namely

It is guaranteed that the histogram h _k has the same distribution degree in the two symmetrical directions during the equalization process.

said for in, is the gray mean value in the segmented histogram h _k .

As shown in Figure 1, it is a schematic diagram of the process of applying the method of the present invention to process images, and the process is as follows:

1) Input the original image I, and calculate the brightness channel histogram of the original image, as shown in Figure (a) and Figure (b);

2) Split the histogram into N segmented histograms, and expand them respectively to obtain scalable segmented histograms As shown in figure (c);

3) For each extended segment histogram Applying the proposed adaptive histogram equalization method, the obtained histogram is shown in Figure (d);

(a) Using the formula Compute the segmented histogram for each extension The weight of , when N=3, corresponds to the schematic diagram of the weight function W ₁ of the segmented histogram h _k , as shown in Figure 3;

(b) Calculate α(k), β(k) and γ(k) using the formulas for optimizing the regularization parameters α, β and γ;

(c) Using the formula $h^u\left(i\right)=\left\{\begin{array}{cc}{W}_k\left(i\right)& i\&NotElement;h_k\\ 1& i\&Element;h_k\end{array}\right.$ Revised uniform distribution histogram H ^u ;

(d) Using the formula Solving for the target histogram The final equalized histogram is calculated, as shown in figure (e), and the final output image is obtained through histogram correction, as shown in figure (f) and figure (g).

In order to verify the effectiveness of the method of the present invention, select different histogram equalization processing methods and the method of the present invention to carry out equalization processing on the same original image, and obtain as shown in Figure 2 (b)-Figure (f) Image enhancement effect comparison chart, from the comparison chart, it can be seen that the grayscale mapping function between the input and output corresponding to figure (f) is obviously better than other methods.

Claims (5)

1. A method based on scalable subsection adaptive histogram equalization, is characterized in that, comprises the following steps: Step 1: Extend the original input histogram using the scalable histogram segmentation method; First, the original histogram H ^o is divided into N non-overlapping segmented histograms h _k , k∈[1,N]; then all segmented histograms h _k are respectively extended to the gray interval of the original histogram H ^o For the same histogram, keep the position of the segmented histogram h _k in the original histogram unchanged, and fill other positions with zeros, that is, an expandable segmented histogram $h_k^x=[0,...,h_k,...,0](k\&Element;[1,N\left]\right);$ Step 2: Equalize the scalable segmented histogram, and each equalized sub-histogram has the same length as the original histogram; Step 3: The equalized sub-histogram obtained in step 2 Perform weighted mixing according to the following formula to obtain the final equalized histogram H ^t (i): ${\mathrm{<span\; class="notranslate">\; h</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>$ where w _k (i) is the normalized weight of the k-th equalized extended segment histogram, weight function k=[1,...,N], N is the number of histogram segments, i is the gray level, u _k and σ _k are the mean and variance of the weight function respectively. 2. based on scalable segmentation adaptive histogram equalization method according to claim 1, it is characterized in that, the mean value of described weight function in, are the first and last gray levels of the segmented histogram h _k respectively; The variance of the weight function To avoid image over-enhancement threshold, is the coverage width of the segmented histogram h _k , namely 3. based on scalable segmentation adaptive histogram equalization method according to claim 2, it is characterized in that, the for in, is the gray mean value in the segmented histogram h _k . 4. according to any one of claim 1-3, based on the scalable segmentation adaptive histogram equalization method, it is characterized in that, the equalization processing selected in the step 2 is adaptive equalization processing, in the present The optimization problem posed by some histogram equalization methods In , the regularization parameters α, β and γ are optimized using the following formulas; $\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; th</span>}}<span\; class="notranslate">\; )</span>$ $\mathrm{<span\; class="notranslate">\; \&beta;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; th</span>}}<span\; class="notranslate">\; )</span>$ γ=1-α-β Among them, H ^x is the expanded histogram of the input image, H ^t is the target histogram of the input image, ^Hu is the expected equalization distribution histogram; H is the histogram obtained after adaptive equalization processing, ||·|| Indicates the Euclidean norm, D is the double diagonal difference matrix; Measure the distance between the gray level i of the expanded histogram H ^x of the input image and the maximum gray level I _max of the expanded histogram H ^x of the input image, i is the gray level in the expanded histogram H ^x of the input image Level, L is the number of gray levels, n _i is the frequency of gray level i in the expanded histogram H ^x of the input image, I _max is the maximum gray level, and the variance σ is 0.2I _max ; m _i is the number of pixels whose gray value I(x _j ) of pixel x _j in the input image is equal to i; nG(x _j ) is the normalized local variance of pixel x _j , and nG(x _j )=G(x _j )/I(x _j ), G(x _j ) is the gradient magnitude at pixel x _j . 5. based on scalable segment adaptive histogram equalization method according to claim 4, it is characterized in that, described expectation equalization distribution histogram ^H adopts following formula to distribute: ${\mathrm{<span\; class="notranslate">\; h</span>}}^{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>& \mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&NotElement;</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\end{array}\right.$ In the formula, i is the gray level.