CN103606137A - Histogram equalization method for maintaining background and detail information - Google Patents

Abstract

The invention relates to the field of digital image processing, and aims at enhancing digital image contrast ratio and weakening unfavorable influence on image quality caused by an over-enhancement effect, a halo effect and average brightness drifting. Therefore, the technical scheme adopted by the invention is a histogram equalization method for maintaining background and detail information. A histogram statistics calculation method for performing statistics on local neighborhood information is adopted so that an optimized histogram is generated. Meanwhile, degree of enhancement of the parameter adjustment contrast ratio is set. Before equalization of the histogram, a certain value of the optimized histogram is taken to perform segmentation, and then histogram equalization operation is performed respectively. The histogram equalization method is mainly applied to digital image processing.

Description

The histogram equalization method that keeps background and detailed information

Technical field

The present invention relates to digital image processing field, the contrast that relates in particular to the gray level image in figure image intensifying strengthens, and extraction and the demonstration of histogram processing and successive image information specifically, relate to the histogram equalization method that keeps background and detailed information.

Background technology

The enhancing technology of digital picture is an importance during image is processed.The interested feature in image can be effectively given prominence in figure image intensifying, weakens less important information, the whole and part information of emphasizing image that can be purposive.For distorted image, the effective Useful Information in Recovery image; The image using for some special occasions, can extract wherein important details.Therefore, no matter figure image intensifying is in military affairs, aviation, geography and commercial use, or can embody its vital role aspect the extraction of infrared image, medical image and photographs and sharpening.

It is a kind of of luminance digital image enhancing that contrast promotes.The contrast that image is good can embody the bright dark difference of image, makes image present abundanter detailed information, has guaranteed better visual effect.Wherein, histogram equalization method is to promote a kind of simple of picture contrast and effective method.

The gray-scale value of histogram equalization method statistic view picture or parts of images, calculates the grey level histogram of image.Obtain afterwards the probability density function that each gray-scale value is corresponding, suppose that image size is for M * N, k gray level number in entire image is n _k, the probability density function of this image gray levels can be obtained by following formula

$\left\{\begin{array}{c}p\left(k\right)=\frac{n_k}{\mathrm{MN}}\\ \underset{k=0}{\overset{255}{\mathrm{\Σ}}}p\left(k\right)=1\end{array}\right.$

By probability density function integration, obtain accumulated probability density function

$c_1\left(k\right)=\underset{s=0}{\overset{k}{\mathrm{\Σ}}}p\left(s\right),k\∈\left[\mathrm{0,255}\right]$

By this accumulated probability density function, can obtain transforming function transformation function

T(k)=255c(k)

The gray-scale value k of each pixel is remapped, can obtain new pixel value.So the contrast of image is improved.When this process is applied on local image, be about to image and be divided into after a plurality of rectangle parts, add up local histogram, and adopt local greyscale transformation function to carry out greyscale transformation to corresponding pixel.More natural for the image that makes to obtain after histogram equalization, some methods are divided into a plurality of parts by histogram, after the first specified dynamic range of every part, carry out the operation of equalization again.The picture contrast and the sharpness that adopt classic method to obtain have had very large lifting, but also can exist such as edge effect, mistake enhancement effect and halation phenomenon.These phenomenons can make image brightness when contrast gets a promotion occur factitious transition, and much noise can obtain unnecessary enhancing, and the visual effect of image can decline, and drift appears in mean flow rate, and distortion appears in image.When adopting partial histogram equalization, the chessboard effect of image is too obvious, has affected the aesthetic property of general image.No matter be to adopt to partly overlap or nonoverlapping image partition method, this effect all cannot be eliminated completely, especially in marginal portion.In addition, statistics with histogram repeatedly has also increased calculated amount.Although the histogram equalization method of dynamic range distribution improves a lot on image visual effect, but still can not be completely free of the drawback that histogram equalization itself brings, inner in each dynamic range, still can have noise and cross the defects such as enhancing, and the heavily distribution of dynamic range can increase calculated amount.Histogrammic equalization method does not all fully take into account the detailed information of image itself above, there is no well to retain the local detail of image, do not overcome histogram equalization cross enhancement effect obviously, the shortcoming such as Noise enhancement and halation result.

Summary of the invention

The present invention is intended to solution and overcomes the deficiencies in the prior art, when promoting image contrast, weakens the adverse effect that enhancement effect, halo effect and average brightness drift are brought picture quality.For this reason, the technical solution used in the present invention is to keep the histogram equalization method of background and detailed information, the statistics with histogram computing method of the local neighborhood information of employing statistics, the histogram that generation is optimized; Setup parameter is adjusted the degree that contrast strengthens simultaneously; Before histogram equalization, the optimization histogram generating is got to certain value and cut apart, then carry out histogram equalization operation respectively.

Further concrete steps are:

First input picture is carried out the statistics of neighborhood information, the gray-scale value of each pixel value of input picture and pixel value around thereof and its relation are calculated, generate two-dimentional optimization histogram;

In the process of neighborhood information statistics, each pixel for input picture, first select its rectangular area of w * w around, w is odd number herein, to all pixels except center pixel in this region, add up the difference between they and center pixel gray-scale value, the gray-scale value of supposing center pixel is m, around the gray-scale value of a certain pixel is n, the coordinate of central pixel point is (i, j), x (i, j) represent its pixel value, the coordinate of surrounding pixel point is (s, t), x (s, t) represent its pixel value, according to following formula

H(m,n)=α|x(i,j)-x(s,t)|+1

What α represented is the enhancing degree of local contrast in neighborhood, in the neighborhood of each rectangle, each pixel will be carried out difference computing with center pixel, w * the w-1 that can calculate in this neighborhood organizes corresponding H value, H is to two-dimensional histogram that should input picture, image for a width M * N, each pixel will be got such neighborhood, through corresponding w * w-1 H value of w * calculate for w-1 time, to produce altogether the individual such H value of M * N * (w * w-1), the H value of same coordinate is added up and just can obtain complete two-dimentional H matrix;

The gray-scale value of all pixels obtains a two-dimentional H matrix through statistics, and this two dimension H matrix is by being converted into the histogram H of one dimension after integration and normalization _m, be exactly the optimization histogram that this input picture is corresponding, two-dimensional matrix is changed into one dimension and optimize histogram H _mformula as follows:

$H_m=\underset{n=0}{\overset{255}{\mathrm{\Σ}}}H(m,n)/\underset{m=0}{\overset{255}{\mathrm{\Σ}}}\underset{n=0}{\overset{255}{\mathrm{\Σ}}}H(m,n);$

After having generated the optimization histogram of input picture, select a suitable numerical value that histogram is divided into two parts subimage X ₁and X ₂, two parts carry out respectively histogram equalization, select this histogrammic area median point or mean point as the threshold value of cutting apart, and establishing revaluate is V _m, according to following formula, can become two independently histogram H by optimizing histogram divion _m1and H _m2:

$\left\{\begin{array}{c}{H}_m=H_{m1}\∪H_{m2}\\ H_{m1}=H_m,m\∈[0,V_m]\\ H_{m2}=H_m,m\∈[V_m+\mathrm{1,255}]\end{array}\right.$

For these two histograms independently, by following formula, obtain accumulated probability density function c separately respectively ₁and c (k) ₂(k):

$\left\{\begin{array}{c}{c}_1\left(k\right)=\underset{s=0}{\overset{k}{\mathrm{\Σ}}}p\left(s\right),k\∈[0,V_m]\\ c_2\left(k\right)=\underset{s=V_M+1}{\overset{k}{\mathrm{\Σ}}}p\left(s\right),k\∈[V_m+\mathrm{1,255}]\end{array}\right.$

Wherein, p (s) is the histogram H optimizing _mcorresponding probability density function, s is corresponding gray-scale value, like this, for gray-scale value in original image, is less than or equal to this V _mpoint, will adopt c ₁(k) as transforming function transformation function, become new pixel value; Gray-scale value is greater than V _mpoint, will adopt c ₂(k) as transforming function transformation function, become new pixel value;

$\left\{\begin{array}{c}{T}_1=V_m{c}_1\left(k\right),k\∈[0,V_m]\\ T_2=V_m+1+[255-(V_m+1)]c_1\left(k\right),k\∈[V_m+\mathrm{1,255}]\end{array}\right.$

After having completed two parts histogram equalization process separately, each pixel has had new corresponding gray-scale value, obtains output image.

The big or small determining step of parameter alpha is: the span of α is [0,4] between, when the value of α is 0, the histogram generating is by consistent with the original histogram of image, in computation process, the value of α be one according to the parameter that needs manual adjustments strengthening, but in the computation process of entire image, the value of α remains unchanged.

A kind of histogram equalization method that keeps background and detailed information that the present invention proposes, in the statistic processes of image histogram, taken into full account the local detail information of image, be included in the probability density function computation process of color histogram, and parameters, well distinguished the enhancing degree of background information and detailed information, when contrast is strengthened, detailed information has also obtained reservation; The method has been protected the mean flow rate of general image simultaneously, makes image before and after processing, not occur that significantly brightness changes; In addition, the method can effectively weaken enhancement effect and halation phenomenon, make image after strengthening not only contrast be improved, also can obtain comparatively naturally visual effect.

The present invention has following beneficial effect:

(1) when the optimization of adding up generation input picture is histogrammic, added up the information of the corresponding neighborhood of each pixel, make the local contrast information that has comprised image in histogram, after the process of histogram equalization in, the detailed information of image can access reservation.

(2) in the process of statistical picture neighborhood information, be provided with manually adjustable parameter, make and to strengthen the size that requires to adjust parameter according to the different characteristics of image, the contrast that image is obtained in various degree promotes, and obtains better visual effect.

(3) after statistics has generated the histogram of optimizing, adopt certain threshold value that histogram is divided into two independently parts, carry out respectively histogram equalization, the mean flow rate that can keep input picture, make the mean flow rate deviation of image after strengthening and input picture less, prevent the distortion that image is excessive.

(4) the method calculated amount of the histogram equalization that the present invention proposes is little, is easy to hardware and realizes, and especially, in consumer electronics product, can meet and calculate soon, the natural requirement of image effect.

Accompanying drawing explanation

Fig. 1 keeps the histogram equalization method flow diagram of background and detailed information.

Fig. 2 neighborhood information statistics schematic diagram.

Fig. 3 Threshold segmentation and histogram equalization schematic diagram.

Embodiment

In order to address the above problem, the present invention, when the histogram of statistical picture, has adopted a kind of statistics with histogram computing method of adding up local neighborhood information, has generated the histogram of optimizing; Setup parameter is adjusted the degree that contrast strengthens simultaneously; Before histogram equalization, the optimization histogram generating is got to certain value and cut apart, then carry out histogram equalization operation respectively.Fig. 1 is the process flow diagram that the present invention carries out histogram equalization method.The concrete grammar of every single stepping will be introduced below.

First the present invention carries out the statistics of neighborhood information to input picture, the gray-scale value of each pixel value of input picture and pixel value around thereof and its relation are calculated, and generates two-dimentional optimization histogram.

In the process of neighborhood information statistics, for each pixel of input picture, first select its rectangular area of w * w around, w is odd number herein.To all pixels except center pixel in this region, add up the difference between they and center pixel gray-scale value, the gray-scale value of supposing center pixel is m, around the gray-scale value of a certain pixel is n, the coordinate of central pixel point is (i, j), x (i, j) represent its pixel value, the coordinate of surrounding pixel point is (s, t), x (s, t) represent its pixel value, according to following formula

H(m,n)=α|x(i,j)-x(s,t)|+1

In the neighborhood of each rectangle, each pixel will be carried out difference computing with center pixel, and the w * w-1 that can calculate in this neighborhood organizes corresponding H value.H is to two-dimensional histogram that should input picture.Image for a width M * N, each pixel will be got such neighborhood, process w * calculate corresponding w * w-1 H value for w-1 time, will produce altogether the individual such H value of M * N * (w * w-1), and the H value of same coordinate is added up and just can obtain complete two-dimentional H matrix.In process in calculating H, need to determine the size of a parameter alpha.What α represented is the enhancing degree of local contrast in neighborhood.The span of α is between [0,4], and α is larger, represents that the high information of local contrast can obtain higher contrast and strengthen, and the information that local contrast is low can obtain lower contrast and strengthen.For some, need the special image that strengthens details, the value of α can slightly obtain larger; For the more images of scenery such as large-area single color and sky, seawater of some background images, the value of α also can slightly obtain larger; Less for other background images, details is not too abundant image also, and the value of α can slightly obtain smaller.When the value of α is 0, the histogram of generation is by consistent with the original histogram of image.In computation process, the value of α be one according to the parameter that needs manual adjustments strengthening.But in the computation process of entire image, the value of α remains unchanged.

After having determined parameter value, the gray-scale value of all pixels just can obtain a two-dimentional H matrix through statistics, and this two dimension H matrix can be converted into the histogram H of one dimension afterwards by integration and normalization _m, be exactly the optimization histogram that this input picture is corresponding.Two-dimensional matrix is changed into one dimension and optimize histogram H _mformula as follows:

$H_m=\underset{n=0}{\overset{255}{\mathrm{\Σ}}}H(m,n)/\underset{m=0}{\overset{255}{\mathrm{\Σ}}}\underset{n=0}{\overset{255}{\mathrm{\Σ}}}H(m,n)$

Histogram after transforming and the difference of original histogram are the local detail information that can better retain image, after carry out in the process of histogram equalization, these information can not lost, the detail section of image also can well be strengthened, and background parts there will not be the transition of sudden change simultaneously.

After having generated the optimization histogram of input picture, select a suitable numerical value that histogram is divided into two parts subimage X ₁and X ₂, two parts carry out respectively histogram equalization, as shown in Figure 3.Can select this histogrammic area median point or mean point as the threshold value of cutting apart.If revaluate is V _m, according to following formula, can become two independently histogram H by optimizing histogram divion _m1and H _m2:

$\left\{\begin{array}{c}{H}_m=H_{m1}\∪H_{m2}\\ H_{m1}=H_m,m\∈[0,V_m]\\ H_{m2}=H_m,m\∈[V_m+\mathrm{1,255}]\end{array}\right.$

For these two histograms independently, can by following formula, obtain accumulated probability density function c separately respectively ₁and c (k) ₂(k):

$\left\{\begin{array}{c}{c}_1\left(k\right)=\underset{s=0}{\overset{k}{\mathrm{\Σ}}}p\left(s\right),k\∈[0,V_m]\\ c_2\left(k\right)=\underset{s=V_M+1}{\overset{k}{\mathrm{\Σ}}}p\left(s\right),k\∈[V_m+\mathrm{1,255}]\end{array}\right.$

Wherein, p (s) is the histogram H optimizing _mcorresponding probability density function, s is corresponding gray-scale value.Like this, for gray-scale value in original image, be less than or equal to this V _mpoint, will adopt c ₁(k) as transforming function transformation function, become new pixel value; Gray-scale value is greater than V _mpoint, will adopt c ₂(k) as transforming function transformation function, become new pixel value.

$\left\{\begin{array}{c}{T}_1=V_m{c}_1\left(k\right),k\∈[0,V_m]\\ T_2=V_m+1+[255-(V_m+1)]c_1\left(k\right),k\∈[V_m+\mathrm{1,255}]\end{array}\right.$

After having completed two parts histogram equalization process separately, each pixel has had new corresponding gray-scale value, obtains output image, and the process of whole histogram equalization has just completed.

Below in conjunction with accompanying drawing and example, the present invention is further illustrated.

The contrast that the present invention is applicable to gray level image strengthens.Suppose that input picture is the 8 bit gradation images of a width 600*800, contain large stretch of sky scenery in this image, color is single, and the more detailed information such as trees house are contained in other regions.First this image is made to neighborhood information statistics, generate the optimization histogram of this image.As shown in Figure 2, first determine the size of rectangular area, select its rectangular area of w * w around, w is 3 herein.To all pixels except center pixel in this region, add up the difference between they and center pixel gray-scale value.The gray-scale value of supposing center pixel is 92, and the gray-scale value of surrounding pixel is for having respectively 92,94,95,96 and 99.

Calculate corresponding H value, also will determine again the size of parameter alpha.What α represented is the enhancing degree of local contrast in neighborhood.The span of α is between [0,4], and α is larger, represents that the high information of local contrast can obtain higher contrast and strengthen, and the information that local contrast is low can obtain lower contrast and strengthen.Due to this image background image that contains large area single color and the part that needs local enhancement details, if so larger α of value, can make the enhancing degree of the part acquisition that local contrast is low lower, can reduce to a certain extent the halation result of background image in enhancing process, make transition more natural; To the scenery such as trees of higher part such as the details of local contrast, can obtain higher enhancing degree, these information can be more clear simultaneously, is easy to identification.Therefore for this image, the value of α chooses 3, and in the computation process of entire image, the value of α remains unchanged.

The value of having determined parameter alpha afterwards just can be according to following formula

H(m,n)=α|x(i,j)-x(s,t)|+1

Calculate around and calculate the H value producing in this neighborhood, it is 92,94 that the pixel that this pixel value is 92 has gray-scale value around, each of 96 pixel, two of the pixels that gray-scale value is 95, three of the pixels that gray-scale value is 99, the H value that they calculate is respectively H (92,92)=1, and H (92,94)=7, H (92,95)=10, H (92,96)=13 and H (92,99)=22.Wherein H (92,95)=10 can be added up twice, H (92,99)=22 meeting quilt statistics three times.These values are just entered in the two-dimensional histogram of this image by statistics, and this two-dimensional histogram is the histogram of 256 * 256 sizes.Like this, for the statistics of single pixel, just completed, the statistics of rest of pixels is carried out in such a manner successively.

After statistics has generated the two-dimensional histogram of input picture, according to formula

$H_m=\underset{n=0}{\overset{255}{\mathrm{\Σ}}}H(m,n)/\underset{m=0}{\overset{255}{\mathrm{\Σ}}}\underset{n=0}{\overset{255}{\mathrm{\Σ}}}H(m,n)$

This two dimension H matrix is by being converted into the histogram H of one dimension after integration and normalization _m, just obtain the optimization histogram that this input picture is corresponding.

Next this histogram is made to Threshold segmentation, select certain value that the histogram divion of this one dimension is become to two independently histograms.Here selecting the threshold point of cutting apart with histogrammic average point conduct, is 112.This be for the histogrammic average brightness value after equalization can be more approaching with the average brightness value of input picture.If revaluate is V _m.Two histograms after cutting apart are respectively H _m1and H _m2:

$\left\{\begin{array}{c}{H}_m=H_{m1}\∪H_{m2}\\ H_{m1}=H_m,m\∈[0,112]\\ H_{m2}=H_m,m\∈[113,255]\end{array}\right.$

For these two histograms independently, can by following formula, obtain accumulated probability density function c separately respectively ₁and c (k) ₂(k):

$\left\{\begin{array}{c}{c}_1\left(k\right)=\underset{s=0}{\overset{k}{\mathrm{\Σ}}}p\left(s\right),k\∈[0,V_m]\\ c_2\left(k\right)=\underset{s=V_M+1}{\overset{k}{\mathrm{\Σ}}}p\left(s\right),k\∈[V_m+\mathrm{1,255}]\end{array}\right.$

Wherein, p (s) is the histogram H optimizing _mcorresponding probability density function, s is corresponding gray-scale value.Like this, in original image, gray-scale value is less than or equal to 112 point, will adopt c ₁(k) as transforming function transformation function, become new pixel value; The point that gray-scale value is greater than 112, will adopt c ₂(k) as transforming function transformation function, become new pixel value.

$\left\{\begin{array}{c}{T}_1=112c_1\left(k\right),k\∈[0,112]\\ T_2=113+(255-113)]c_1\left(k\right),k\∈[\mathrm{113,255}]\end{array}\right.$

Like this, for each width input picture, all can obtain two corresponding transforming function transformation functions, each pixel value can utilize one of them transforming function transformation function to obtain corresponding new pixel value.After completing this whole step, just can generate the output image that a width is new, the contrast of this image can be significantly improved.Meanwhile, it is even that the brightness at background position changes nature, and the details of image also can well be retained.The process of whole histogram equalization has just completed.

Claims (3)

1. a histogram equalization method that keeps background and detailed information, is characterized in that, adopts the statistics with histogram computing method of the local neighborhood information of statistics, generates the histogram of optimizing; Setup parameter is adjusted the degree that contrast strengthens simultaneously; Before histogram equalization, the optimization histogram generating is got to certain value and cut apart, then carry out histogram equalization operation respectively.

2. the histogram equalization method of maintenance background as claimed in claim 1 and detailed information, is characterized in that, further concrete steps are:

First input picture is carried out the statistics of neighborhood information, the gray-scale value of each pixel value of input picture and pixel value around thereof and its relation are calculated, generate two-dimentional optimization histogram;

In the process of neighborhood information statistics, each pixel for input picture, first select its rectangular area of w * w around, w is odd number herein, to all pixels except center pixel in this region, add up the difference between they and center pixel gray-scale value, the gray-scale value of supposing center pixel is m, around the gray-scale value of a certain pixel is n, the coordinate of central pixel point is (i, j), x (i, j) represent its pixel value, the coordinate of surrounding pixel point is (s, t), x (s, t) represent its pixel value, according to following formula

H(m,n)=α|x(i,j)-x(s,t)|+1

In the neighborhood of each rectangle, each pixel will be carried out difference computing with center pixel, w * the w-1 that can calculate in this neighborhood organizes corresponding H value, H is to two-dimensional histogram that should input picture, for the image of a width M * N, each pixel will be got such neighborhood, through corresponding w * w-1 H value of w * calculate for w-1 time, to produce altogether the individual such H value of M * N * (w * w-1), the H value of same coordinate is added up and just can obtain complete two-dimentional H matrix;

The gray-scale value of all pixels obtains a two-dimentional H matrix through statistics, and this two dimension H matrix is by being converted into the histogram H of one dimension after integration and normalization _m, be exactly the optimization histogram that this input picture is corresponding, two-dimensional matrix is changed into one dimension and optimize histogram H _mformula as follows:

$H_m=\underset{n=0}{\overset{255}{\mathrm{\Σ}}}H(m,n)/\underset{m=0}{\overset{255}{\mathrm{\Σ}}}\underset{n=0}{\overset{255}{\mathrm{\Σ}}}H(m,n);$

After having generated the optimization histogram of input picture, select a suitable numerical value that histogram is divided into two parts subimage X ₁and X ₂, two parts carry out respectively histogram equalization, select this histogrammic area median point or mean point as the threshold value of cutting apart, and establishing revaluate is V _m, according to following formula, can become two independently histogram H by optimizing histogram divion _m1and H _m2:

$\left\{\begin{array}{c}{H}_m=H_{m1}\∪H_{m2}\\ H_{m1}=H_m,m\∈[0,V_m]\\ H_{m2}=H_m,m\∈[V_m+\mathrm{1,255}]\end{array}\right.$

For these two histograms independently, by following formula, obtain accumulated probability density function c separately respectively ₁and c (k) ₂(k):

$\left\{\begin{array}{c}{c}_1\left(k\right)=\underset{s=0}{\overset{k}{\mathrm{\Σ}}}p\left(s\right),k\∈[0,V_m]\\ c_2\left(k\right)=\underset{s=V_M+1}{\overset{k}{\mathrm{\Σ}}}p\left(s\right),k\∈[V_m+\mathrm{1,255}]\end{array}\right.$

Wherein, p (s) is the histogram H optimizing _mcorresponding probability density function, s is corresponding gray-scale value, like this, for gray-scale value in original image, is less than or equal to this V _mpoint, will adopt c ₁(k) as transforming function transformation function, become new pixel value; Gray-scale value is greater than V _mpoint, will adopt c ₂(k) as transforming function transformation function, become new pixel value;

$\left\{\begin{array}{c}{T}_1=V_m{c}_1\left(k\right),k\∈[0,V_m]\\ T_2=V_m+1+[255-(V_m+1)]c_1\left(k\right),k\∈[V_m+\mathrm{1,255}]\end{array}\right.$

After having completed two parts histogram equalization process separately, obtain output image.

3. the histogram equalization method of maintenance background as claimed in claim 1 and detailed information, is characterized in that, the big or small determining step of parameter alpha is: what α represented is the enhancing degree of local contrast in neighborhood.The span of α is between [0,4], when the value of α is 0, the histogram generating is consistent with the original histogram of image, in computation process, the value of α be one according to the parameter that needs manual adjustments strengthening, but in the computation process of entire image, the value of α remains unchanged.

Images