CN104268843A - Image self-adaptation enhancing method based on histogram modification - Google Patents

Abstract

The invention discloses an image self-adaptation enhancing method based on histogram modification. Firstly, the standard deviation of an original histogram is calculated, and the original histogram and the standard deviation of the original histogram are added to obtain a primarily-corrected histogram; secondly, the self-adaptive Gamma correction is carried out on the basis of the primarily-corrected histogram, and a secondarily-corrected histogram is obtained; finally, a target-enhanced image is obtained through a traditional histogram enhancement method on the basis of the secondarily-corrected histogram. The image self-adaptation enhancing method has the advantages that adaptability is high, and the method is suitable for self-adaptation enhancement of various images; the information loss is low in the enhancement processing process, the details of the images can be effectively kept, and most of detail information of original images can be reserved in the enhanced images obtained through the method; the excessive changing of the image brightness is avoided, the original characteristics of the various images can be effectively reserved, and the excessive changing of the image brightness is avoided.

Description

Based on the image self-adapting enhancement method that histogram is modified

Technical field

The present invention relates to image processing techniques, particularly a kind of image self-adapting enhancement method based on histogram modification.

Background technology

Picture superposition is an important technology in visually-perceptible and machine vision, is widely used in Medical Image Processing, video monitoring system, in the systems such as satellite image process.The target of contrast strengthen improves picture contrast, provides directly perceived, clear, is suitable for the image analyzed.Numerous method for enhancing picture contrast can be classified as two classes substantially: directly strengthen and indirectly strengthen.Direct Enhancement Method refers to the measurement standard of certain token image contrast of definition, and algorithm for design strengthens this variable of image; Indirect method then strengthens picture contrast by the mode redistributing former gray level.Indirect method is owing to having the advantage such as effective, directly perceived and being paid close attention to widely.It can be further subdivided into: the method based on gray scale transformation and the method based on histogram treatment.Histogram equalization is one of a kind of quick, effective, classical image enchancing method based on histogram treatment.It is using original histogram as input, former histogrammic cumulative distribution function is utilized to generate mapping function, original narrow grey level range is mapped to a wider grey level range, to increase the dynamic range of image gray levels, reaches the object strengthening image.Although histogram equalization has fast, efficient, be easy to the advantages such as realizations, have that details is easily lost also, significantly brightness change, the obvious defect such as luminance saturation and stereovision difference.

For these defects, many documents have been had to propose some different solutions.

The people such as Arici propose a kind of framework of histogram modification and image enchancing method thereof, and (details are see document: T.Arici, S.Dikbas, and Y.Altunbasak, " A histogram modification framework and its application for image contrast enhancement. " IEEE Trans.Image Process., Vol.18, no.9, pp.1921-1934, Sep.2009.).The excessive enhancing that they avoid smooth region by introducing new statistics with histogram mode; Histogram is revised, to avoid the problems such as loss in detail by introducing weighted mean; And stretch in conjunction with black and white with the dynamic range making full use of gray level.The method improves a lot on the basis of traditional histogram equalization, but fundamentally can not avoid the defect of traditional histogram equalization method, and just degree decreases.

The people such as Huang propose the histogram enhancement method distributed based on self-adaptation Gamma correction and weighted, and (details are see document: S.C.Huang, F.C.Cheng, and Y.S.Chiu, " Efficient contrast enhancement using adaptive gamma correction with weighting distribution. " IEEE Trans.Image Process., Vol.22, no.3, pp.1032-1041, Mar.2013.).The method utilizes self-adaptation Gamma correction to strengthen image low-light level part and to suppress the obvious reduction of image highlighted part contrast; Utilize weighted to distribute and revise former histogram, avoid shortcoming when applying traditional histogram equalization.Although the method effectively can avoid the defect of traditional histogram equalization, there is grayscale dynamic range and utilize insufficient, strengthen the problems such as DeGrain.

Summary of the invention

The object of the invention is to overcome above-mentioned the deficiencies in the prior art, a kind of image self-adapting enhancement method based on histogram modification is provided, this method can not only make image detail and the brightness of maintenance, and under illumination condition complicated and changeable, image energy self-adaptation strengthens.

The technical scheme realizing the object of the invention is:

Based on an image self-adapting enhancement method for histogram modification, comprise the following steps:

(1) with programming tool (opencv, matlab etc.) read the digital picture needing to strengthen, obtain the matrix expression { f (i of target image, j) }, wherein f (i, j) represents image { f (i, j) any one pixel }, i, j are the horizontal ordinate and ordinate that pixel f (i, j) is corresponding;

(2) histogram h (k) of computed image { f (i, j) }, wherein h (k) represents gray-scale value is the frequency that the pixel of k occurs in the target image;

(3) the standard deviation δ of the middle nonzero element of compute histograms h (k);

(4) once revised histogram h is calculated ₁(k), wherein h ₁k ()=h (k)+δ, h (k) is original histogram, h ₁k () is for revising rear histogram;

(5) computed image average brightness value μ;

(6) adaptive gamma correction coefficient gamma is calculated, wherein $\mathrm{\&gamma;}=\left\{\begin{array}{cc}(255-\mathrm{\&mu;})/255& \mathrm{\&mu;}<128\\ \mathrm{\&mu;}/255& \mathrm{\&mu;}\&GreaterEqual;128\end{array}\right.;$

(7) the histogram h after second-order correction is calculated ₂(k), wherein h ₂(k)=[h ₁(k)] ^γ;

(8) (7) revised histogram h is utilized ₂k (), application histogram equalization formula carries out histogram equalization.

The calculating formula of described step (3) Plays difference δ is:

$\mathrm{\&delta;}=\sqrt{\frac{1}{K}\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}{[\tilde{h}\left(k\right)-u]}^2},$

In formula, K is total number of greyscale levels in image, for the set of nonzero element in former histogram h (k), u is average, its calculating formula is:

$u=\frac{\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}\tilde{h}\left(k\right)}{K}.$

In described step (5), the computing formula of mean picture brightness value μ is:

$\mathrm{\&mu;}=\frac{\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}(\tilde{h}\left(k\right)\&times;k)}{N}\text{.}$

In formula, N is total number of pixel in image { f (i, j) }.

In described step (8), the computing method of histogram equalization utilize formulae discovery mapping function: in formula, T (k) represents the mapping value of gray-scale value k after this method, and cdf (k) for the cumulative distribution function after histogram modification, its calculating formula is:

wherein k=0,1 ..., K-1.

Here p (k) is the probability density function after histogram modification, and its calculating formula is:

wherein k=0,1 ..., K-1.

The present invention, compared with the existing method based on histogram enhancement, has the following advantages: 1) adaptability is good: can adaptive process generate the image that high-quality strengthens for excessively bright, excessively dark, the balanced image of entirety; 2) balanced enhancing: can strengthen by efficient balance image each several part, there will not be and enhance image part region contrast, but reduce the phenomenon of other a part of contrast; 3) effectively retain characteristics of image: the present invention efficiently can strengthen image and keep image detail information and mean flow rate, avoid luminance saturation, brightness significantly to change and loss in detail.4) oeverall quality is high: computation complexity of the present invention is low, and adaptability is good, effectively can take into account the maintenance of image enhaucament and image detail, brightness etc.

Accompanying drawing explanation

In accompanying drawing, Fig. 1-6 is the experimental result of embodiment 1.Wherein Fig. 1 is tiffany former figure, Fig. 2 is traditional histogram treatment result, and Fig. 3 is the result of the method that the people such as Arici propose, and Fig. 4 is the result of the method that the people such as Huang propose, and Fig. 5 is the result of the inventive method; The original histogram that Fig. 6 (a) is Fig. 1, the histogram of Fig. 6 (b) corresponding to Fig. 2, the histogram of Fig. 6 (c) corresponding to Fig. 3, the histogram of Fig. 6 (d) corresponding to Fig. 4, the histogram of Fig. 6 (e) corresponding to Fig. 5.

In accompanying drawing, Fig. 7-12 is the experimental result of example 2.Wherein Fig. 7 is Hawkes_Bay former figure, Fig. 8 is traditional histogram treatment result, and Fig. 9 is the result of the method that the people such as Arici propose, and Figure 10 is the result of the method that the people such as Huang propose, and Figure 11 is the result of the inventive method; The original histogram that Figure 12 (a) is Fig. 7, the histogram of Figure 12 (b) corresponding to Fig. 8, the histogram of Figure 12 (c) corresponding to Fig. 9, the histogram of Figure 12 (d) corresponding to Figure 10, the histogram of Figure 12 (e) corresponding to Figure 11.

In accompanying drawing, Figure 13-18 is the experimental result of example 3.Wherein Figure 13 is Hawkes_Bay former figure, Figure 14 is traditional histogram treatment result, and Figure 15 is the result of the method that the people such as Arici propose, and Figure 16 is the result of the method that the people such as Huang propose, and Figure 17 is the result of the inventive method; The original histogram that Figure 18 (a) is Figure 13, the histogram of Figure 18 (b) corresponding to Figure 14, the histogram of Figure 18 (c) corresponding to Figure 15, the histogram of Figure 18 (d) corresponding to Figure 16, the histogram of Figure 18 (e) corresponding to Figure 17.

Embodiment

For further illustrating content of the present invention, below in conjunction with specific embodiments and the drawings, content of the present invention and embodiment and superiority are elaborated.Wherein embodiment 1 is the partially bright example of integral image, and embodiment 2 is that integral image brightness is moderate, but concentrates on the example in the middle part of gray level, and embodiment 3 is the partially dark example of integral image.

Embodiment 1:

The present embodiment is that experimental study instrument is to illustrate concrete implementation step with Matlab.Experimental subjects Fig. 1 gets [0,255] the tiffany image (note: tiffany image takes from American South University of California signal and image procossing research institute USC-SIPI image data base in tonal range, the image data base that this storehouse is image procossing, graphical analysis and field of machine vision are extensively quoted, the network address of quoting of this image library is: http://sipi.usc.edu/database/).This image slices vegetarian refreshments mainly concentrates in the intensity value ranges of [150 255], and entirety is partially bright, and the concrete implementation step that image tiffany strengthens is as follows:

1. image reading function imread is utilized to read in tiffany image, the statement reading in tiffany image is: img=imread (' tiffany.GIFf '), obtain the memory variable img of target image, img (i, j) internal memory corresponding to the arbitrary pixel of image represents, i, j are the subscript that pixel internal memory represents img (i, j).

2. utilize matlab statistics with histogram function imhist computed image histogram imgHist, its statement is: imgHist=imhist (img), and the internal memory obtaining image histogram expresses imgHist.Because image grayscale range is [0,255], so variable i mgHist contains 256 elements, each element value imgHist (k) represents the frequency that corresponding grey scale value k-1 occurs in the picture.

3. the standard deviation delta of matlab function std compute histograms imgHist is utilized.Its statement is: delta=std (imgHist).

4. utilize on former histogram and add its standard deviation delta, calculate once revised histogram imgHist1:imgHist1=imgHist+delta.

5. computed image mean flow rate miu.The statement using the mean value computation function mean of matlab instrument to calculate mean flow rate is: miu=mean (mean (img)).

6. utilize above in summary of the invention step (6) expression formula calculate gamma correction coefficient gamma of the present invention.

7. calculate the histogram imgHist2 after second-order correction, its matlab statement is: imgHist2=(imgHist1) ^gamma.

8. after second-order correction histogram imghist2 basis on, apply the method for traditional histogram equalization, carry out equalization processing, the mapping method of this process is: tk=round ((256-1) × cdf (k)), here the round function that provides for matlab instrument of round.Namely in former figure, gray-scale value is the pixel of k, and after the inventive method process, gray-scale value is tk.

For the superiority of the inventive method in image enhaucament oeverall quality is described, by the method that the inventive method proposes for people such as the result of tiffany image and Arici, the method that the people such as Huang propose and traditional histogram equalization method result compare, experimental result is as Fig. 1 of patent specification accompanying drawing of the present invention, Fig. 2, shown in Fig. 3, Fig. 4, Fig. 5 and Fig. 6.

As can be seen from the original image of Fig. 1 and the original histogram of Fig. 6 (a), tiffany imaging surface has vaporific covering, and brightness entirety is partially bright.From Fig. 2, Fig. 3, the result of Fig. 4 and Fig. 5 can be found out, all there is the problem excessively strengthened in the method except the inventive method: changes into grey black after tiffany gray background behind strengthens, part hair near tiffany auris dextra has also occurred that larger brightness changes, the forehead of Tiffany, eyebrow, factitious visual impression has been there is in eyeball because excessive brightness changes, then there are not the problems referred to above in result Fig. 5 of the inventive method, this figure is clear, natural as seen from Figure 5, and details is enriched.

For further superiority of the present invention being objectively described, the special discrete entropy (DE) introducing image, contrast variable (MEME), absolute difference (AMBE) three indexs of mean flow rate are next objective portrays each method performance.

Image discrete entropy feature piece image comprise the number of quantity of information: the discrete entropy of image is larger, and key diagram picture comprises more detailed information.

Due to the asymmetry of information processing, the output entropy of any image conversion all can not exceed its input entropy.The discrete entropy of image is defined as:

$\mathrm{DE}=-\underset{k=0}{\overset{K}{\mathrm{\&Sigma;}}}{p}_i\left(k\right)\log p_i\left(k\right),$

P in formula _ik () is the probability density of the pixel of k for gray-scale value in input picture.

What contrast variable characterized is the size of picture contrast: to same sub-picture, contrast variable is larger, represents that this picture contrast is larger, but and does not mean that visual quality of images is higher.Assuming that image I is divided into k ₁× k ₂individual size is M ₁× M ₂fritter (i, j) is each fritter center, the definition of contrast variable (MEME) is:

$\mathrm{MEME}=\mathrm{\&alpha;}\&times;\frac{C_{I_{\mathrm{DC}}}}{k_1\&times;k_2}\&times;\frac{1}{k_1\&times;k_2}\underset{i=1}{\overset{k_1}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{k_2}{\mathrm{\&Sigma;}}}{C}_{i,j}^w,$

I in formula _dCby the k of I ₁× k ₂the thumbnail that the DC component of individual image block is formed, at I _dCin the gray-scale value of each pixel and k ₁× k ₂in the average of all pixels of certain image block corresponding; α is an amplification regulating parameter, avoids net result numerical value too little (α is set as 100 in the present invention). with be root mean square contrast, it is defined as:

$C=\sqrt{\frac{1}{L\&times;M}\underset{m=1}{\overset{L}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{M}{\mathrm{\&Sigma;}}}{(I_{m,n}-\mathrm{mean}\left(I_{m,n}^w\right))}^2}.$

The absolute difference of input and output image averaging gray scale that what the absolute difference (AMBE) of mean flow rate was measured is.The absolute difference of mean flow rate is less, and illustrate that the average gray of original image keeps better, the definition of the absolute difference (AMBE) of mean flow rate is:

AMBE＝|u _i-u _o|，

In formula, u _iand u _oinput value and the output valve of average gray in image enhancement processes respectively.

As can be seen from Table 1, the highest image discrete entropy is achieved to tiffany image method of the present invention, illustrate that the present invention remains the detailed information of former figure preferably.From contrast variable, the present invention does not occur excessively strengthening phenomenon, and the change of the mean flow rate of the inventive method is minimum, and much smaller than other reference methods, the visual effect of Fig. 5 also verifies above-mentioned conclusion.

Table 1

The objective evaluation of Tiffany processing result image quality

Embodiment 2:

Experimental subjects Fig. 7 of embodiment 2 gets [0,255] Hawke ' the s bay image (note: this image takes from histogram equalization entry figure in English wikipedia wikipedia in tonal range, this image is by Phillip Capper in the shooting of New Zealand Hawke's Bay area, and image network is linked as: http://commons.wikimedia.org/wiki/File%3AHawkes_Bay_NZ.jpg).This pixel point concentrates in the narrow range of [120,200], and original Contrast's degree is very low, has and covers mist sense.The experimental result of embodiment 2 as illustrated in figs. 8-11, the histogram of Figure 12 (a)-(e) corresponding to Fig. 8-11.

From the experimental data of table 2, the absolute difference of the mean flow rate of traditional histogram equalization method is minimum, and contrast strengthen is maximum, but the image visual effect of Fig. 8 is but the poorest.This is because original histogram Figure 12 (a) concentrates on the middle part of gray level, traditional histogram equalization method stretches overall for this gray level to two ends, thus obtains larger enhancing while maintenance mean picture brightness.But due to the inherent shortcoming of traditional histogram equalization method when processing unbalanced image, Fig. 8 lost a lot of image detail information, the small river divided as original image middle and upper part becomes and can distinguish hardly after enhancing, the hillside of image middle and lower part is then a vast expanse of whiteness, and in table 2, the Image entropy of traditional histogram equalization method is minimum has also confirmed above-mentioned conclusion.

Method treatment effect and traditional histogram equalization method of people's propositions such as Arici are similar, are only that shortcoming degree weakens to some extent.Although the method that the people such as Huang propose maintains the small river information of dividing former figure middle and upper part preferably, but lost the detailed information of image middle and lower part grove.From the histogram of Figure 12 (d), the center of former Nogata about Figure 150 is moved to the position of about 50 by the method that the people such as Huang propose, the excessive brightness not only causing as shown in table 2 90.16 changes, also have compressed the grey level range compared with dark pixel, thus cause the loss compared with dark-part details in original image.

And method of the present invention keeps on former figure details basis to the full extent, take into account the absolute difference of contrast variable and mean flow rate preferably, keep and enhance the detailed information of small river and the image middle and lower part grove of dividing image middle and upper part, it also avoid the excessive blast problem of the hill portion of image middle and lower part.It exports, and the entropy of not only image is high, detailed information complete, and brightness changes moderate, and stereovision is strong, good visual effect.

Table 2

The objective evaluation of Hawke ' s bay processing result image quality

Embodiment 3:

Experimental subjects Figure 13 of embodiment 3 gets log cabin image in [0,255] tonal range (note: this image is the log cabin of the ethnic group that the present invention first inventor takes in Guangxi) within the border.As shown in figure 13, this figure prospect background is clearly demarcated, and prospect lamp housing portion is limpid in sight, and background log cabin inside is dim.As shown in Figure 18 (a), the histogram main body of this figure is distributed in the narrow range of [030], but also has small part pixel to be dispersed in the interval of [100 255].This image pixel skewness, the boundary of light and shade part is obvious, and image itself has larger background noise, is the typical image that checking image strengthens quality.The experimental result of embodiment 3 as shown in figures 14-17, the histogram of Figure 18 (a)-(e) corresponding to Figure 13-17.

Experimental result as can be seen from Figure 14 and table 2: traditional histogram equalization method achieves the highest contrast variable and maximum mean flow rate absolute difference, but output is a secondary image excessively strengthened.Brightness of image has larger change, and entirety is brighter, but noise information is also greatly strengthened, and causes image detail to lose, and sharpness reduces.And no matter the method that the people such as Huang proposes is in the absolute length chang image entropy experimental data of contrast variable, mean flow rate, or in the output image visual effect of Figure 16 with former figure all without obviously changing.

The result of people's put forward the methods such as Arici is close with the inventive method on the absolute length chang image entropy of contrast variable, mean flow rate, but not as the inventive method in visual effect.The image that the inventive method strengthens is brighter, and equilibrium enhances the information at log cabin dark place, and the method that the people such as Arici proposes can not details in complete displaying log cabin.On the other hand, on the method that the people such as Arici proposes is fuzzy image lower right corner window through speck, method of the present invention is complete remains brighter speck information.

Table 3

The objective evaluation of log cabin processing result image quality

Claims (4)

1. based on an image self-adapting enhancement method for histogram modification, it is characterized in that, comprise the following steps:

(1) digital picture needing to strengthen is read with programming tool, obtain the matrix expression { f (i of target image, j) }, wherein f (i, j) any one pixel in image { f (i, j) } is represented, i, j is the horizontal ordinate and ordinate that pixel f (i, j) is corresponding;

(2) histogram h (k) of computed image { f (i, j) }, wherein h (k) represents gray-scale value is the frequency that the pixel of k occurs in the target image;

(3) the standard deviation δ of the middle nonzero element of compute histograms h (k);

(4) once revised histogram h is calculated ₁(k), wherein h ₁k ()=h (k)+δ, h (k) is original histogram, h ₁k () is for revising rear histogram.

(5) computed image average brightness value μ;

(6) adaptive gamma correction coefficient gamma is calculated, wherein $\mathrm{\&gamma;}=\left\{\begin{array}{cc}(255-\mathrm{\&mu;})/255& \mathrm{\&mu;}<128\\ \mathrm{\&mu;}/255& \mathrm{\&mu;}\&GreaterEqual;128\end{array}\right.;$

(7) the histogram h after second-order correction is calculated ₂(k), wherein h ₂(k)=[h ₁(k)] ^γ;

(8) (7) revised histogram h is utilized ₂k (), application histogram equalization formula carries out histogram equalization.

2. image self-adapting enhancement method according to claim 1, is characterized in that: the calculating formula of described step (3) Plays difference δ is:

$\mathrm{\&delta;}=\sqrt{\frac{1}{K}\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}{[\tilde{h}\left(k\right)-u]}^2},$

In formula, K is total number of greyscale levels in image, for the set of nonzero element in former histogram h (k), u is average, its calculating formula is:

$u=\frac{\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}\tilde{h}\left(k\right)}{K}.$

3. image self-adapting enhancement method according to claim 1, is characterized in that: in described step (5), the computing formula of mean picture brightness value μ is:

$\mathrm{\&mu;}=\frac{\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}(\tilde{h}\left(k\right)\&times;k)}{N}.$

In formula, N is total number of pixel in image { f (i, j) }.

4. image self-adapting enhancement method according to claim 1, is characterized in that: in described step (8), the computing method of histogram equalization utilize formulae discovery mapping function: wherein T (k) represents the mapping value of gray-scale value k after this method, and cdf (k) for revised cumulative distribution function, its calculating formula is:

$C\left(k\right)=\underset{l=0}{\overset{k}{\mathrm{\&Sigma;}}}p\left(k\right),$ Wherein k=0,1 ..., K-1.

Here p (k) is revised probability density function, and its calculating formula is:

$p\left(k\right)=\frac{h_2\left(k\right)}{\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}{h}_2\left(k\right)}$ Wherein k=0,1 ..., K-1.