CN104463806A - Highly adaptive image contrast enhancing method based on data driving technology - Google Patents

Abstract

The invention discloses a content-based block adaptive image contrast enhancing algorithm to solve the problem that adaptive processing of various degraded images with different characteristics can not be achieved in the prior art by the adoption of the data driving idea. According to the method, by means of block analysis and processing, images are better processed by means of the details and local information of the images; a parameterized enhancement function is established and can change the characteristic of an enhancement curve by adjusting relevant parameters, and then a corresponding enhancement function curve is generated for each image according to the characteristic of the image; features relevant to the enhancement function are extracted through content analysis of image subblocks, and matched enhancement parameters are automatically generated according to the features and assigned to the enhancement function to enable the features of the images to be organically related with the characteristics of the enhancement function. By the adoption of the method, adaptive processing of various degraded images with different characteristics can be achieved without manual intervention.

Description

Based on the height adaptive method for enhancing picture contrast of data driven technique

Technical field

The invention belongs to areas of information technology, particularly a kind of height adaptive method for enhancing picture contrast based on data driven technique.

Background technology

Contain very abundant information in image, in all information of mankind's acceptance, visual information accounts for 80%, so image is very important information delivery media and mode according to statistics.Therefore, be all focus for a long time to the research of image procossing.Image enhaucament strengthens the useful information in image, it can be the process of a distortion, its objective is the visual effect will improving image, for the application scenario of Given Graph picture, on purpose emphasize entirety or the local characteristics of image, original unsharp image is become clear or emphasizes some interested feature, difference in expanded view picture between different objects feature, suppress uninterested feature, thus improve picture quality, abundant information amount, strengthen image interpretation and recognition effect, meet the needs of some special analysis.Have very important meaning to the research of video image enhancement, it not only can as an independently system, the video image information of outputting high quality; Also can as the preprocessing subsystem of many semantic level algorithms, the low-quality image originally not meeting algorithm requirement is carried out processing and is met the high quality graphic that algorithm requires, to ensure the validity of subsequent algorithm.

Image enhaucament has very important significance in image processing field, is for many years one of the study hotspot in this field always.Picture superposition exists usually used as the pretreatment module of image processing system, and its input is generally that contrast is lower, is difficult to the original image meeting subsequent algorithm requirement; Its output is then that contrast is promoted, and reaches the enhancing result that subsequent algorithm requires; Contrast enhancement algorithms is exactly the process by certain brightness correction function correction original gray value.Existing picture superposition algorithm is roughly divided into two classes, and a class is non-self-consistent method, as histogram equalization algorithm, histogram specification method etc., the shortcoming of these class methods is: be usually only applicable to a certain class image, broad applicability is poor, is difficult to be promoted.Another kind of is adaptive approach, adaptive contrast enhancement algorithms core is just, it can by analyzing the feature of image, extract, and automatic generation is applicable to the rectification function with these characteristic images according to this, thus realize accurately processing targetedly the carrying out of different images intelligence.The algorithm for image enhancement possessing adaptive ability gets more and more, comprise the contrast enhancement algorithms of the low bright algorithm for image enhancement of content-based hyperchannel, weight allocation self-adaptation Gamma correction, but these algorithms remain in some problems, such as strengthen low effort, not good to the complicated image treatment effect of mixing multifrequency nature, adaptability scope is less.

Summary of the invention

The object of the present invention is to provide a kind of contrast enhancement process that can be widely used in multiple different qualities degraded image, obtain desirable enhancing result.

The technical scheme realizing the object of the invention is: a kind of adapting to image contrast enhancement process based on data-driven, comprises the following steps:

Step 1, image is transformed into form and aspect, colourity, gray scale color space from three primary colors color space; Formula used is:

v＝max

Wherein, r, g, b are respectively the value of pixel red, green, blue passage, and max is equivalent to the maximum in r, g, b, and min equals the reckling in these values, and h is the angle value of chrominance space, and value set standardization is between 0 to 360 °.

Step 2, in gray channel, adopt the window of M × N size to travel through with the sliding window method that sliding step is 1 pixel image, wherein M, N are respectively line number and the columns of image subblock; Be specially:

Step 2-1, the subimage in window is carried out to the analysis of gray-scale value and contrast, first, obtain subimage grey level histogram, formula used is:

$\mathrm{pdf}\left(l\right)=\frac{n_l}{M\×N},$

Wherein, l (l ∈ [0,255]) is gray level, n _lbe the number of the pixel of l for gray scale, M, N are respectively line number and the columns of image subblock;

Then, its gray average and contrast is calculated according to image grey level histogram:

$\mathrm{avg}=\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}[\frac{l}{l_{\max }}\·\mathrm{pdf}\left(l\right)],$

$\mathrm{con}\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}[{(l-\mathrm{avg})}^2\·\mathrm{pdf}\left(l\right)]$

Wherein, l _maxit is the maximal value of l;

Step 2-2, according to the grey level histogram of step 2-1 neutron image and the contrast of gray channel, set up the parameterized gray probability distribution function of subimage; Be specially:

${\mathrm{pdf}}^{\′}\left(l\right)={\mathrm{pdf}}_{\mathrm{mac}}\·{\left(\frac{\mathrm{pdf}\left(l\right)-{\mathrm{pdf}}_{\min }}{{\mathrm{pdf}}_{\max }-{\mathrm{pdf}}_{\min }}\right)}^{(1+\sqrt{\mathrm{con}})}$

Wherein, pdf _minand pdf _maxfor the minimum value in each element of grey level histogram vector and maximal value;

The subimage parametrization gray probability distribution function that step 2-3, the eigenwert of subimage extracted according to step 2-1 and step 2-2 obtain, comprehensively goes out to be applicable to the enhancing function of this subimage, is specially:

$\mathrm{cf}\left(l\right)=l_{\max }{\left(\frac{l}{l_{\max }}\right)}^{1+\mathrm{cv}\·\sqrt{\mathrm{gn}\left(l\right)}}$

Wherein, $\mathrm{gn}\left(l\right)=\underset{l=1}{\overset{l}{\mathrm{\Σ}}}\frac{{\mathrm{pdf}}^{\′}\left(l\right)}{\mathrm{sum}\left({\mathrm{pdf}}^{\′}\right)},\mathrm{cv}=\left\{\begin{array}{cc}1,& (\mathrm{avg}\≥180)\\ -1,& \left(\mathrm{else}\right)\end{array}\right.;$ Wherein, sum () is the summation operation to vectorial all elements;

Step 2-4, the gray-scale value of pixel each in subimage carried out to enhancing process, formula used is:

gray'(i,j)＝cf[gary(i,j)]

Wherein, gray (i, j) is original gray value, gray'(i, j) be the gray-scale value after enhancing, (i, j) is the coordinate of current pixel point in view picture figure.

Step 3, the result gray-scale value of sliding window method traversal gray channel to be normalized, the gray channel image after being enhanced; Being normalized formula used to the result gray-scale value of sliding window method traversal gray channel is:

${\mathrm{gray}}^{\′}(i,j)=\frac{{\mathrm{\Σ}}_{n=1}^{M\×N}G\left(n\right)}{M\×N}$

Wherein, G (n) to be processed the set of the gray-scale value obtained for each pixel by M × N time.

Step 4, the chrominance channel of image to be revised; Specifically carry out revising according to the grey scale change degree before and after strengthening, described strengthen before and after grey scale change degree be:

$\mathrm{dg}=\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}\frac{l}{l_{\max }}[{\mathrm{pdf}}_e\left(l\right)-{\mathrm{pdf}}_o\left(l\right)]$

Wherein, pdf _eand pdf _obe respectively the grey level histogram of image before and after strengthening, revised chrominance representation is:

C (s)=s ^1-dg, wherein s is the chromatic value of each pixel.

Step 5, the gray channel after strengthening to be merged with revised chrominance channel and original form and aspect passage, and be again transformed into rgb color space, the coloured image after being finally enhanced.Formula used is:

$f=\frac{h}{60}-h_i$

p＝v×(1-s)

q＝v×(1-f×s)

t＝v×(1-(1-f)×s)

$(r,g,b)=\begin{array}{c}\left\{\begin{array}{c}(v,t,p),\mathrm{if}{h}_i=0\\ (q,v,p),\mathrm{if}{h}_i=1\\ (p,v,t),\mathrm{if}{h}_i=2\\ (p,q,v),\mathrm{if}{h}_i=3\\ (t,p,v),\mathrm{if}{h}_i=4\\ (v,p,q),\mathrm{if}{h}_i=5\end{array}\right.\end{array}$

Wherein, h is form and aspect, and s is colourity, and v is brightness, and r is red color component value, and g is green component values, and b is blue color component value, h _iby to h divided by 60 business round mould 6 downwards and try to achieve.

The present invention compared with prior art, has the following advantages:

(1) the present invention is owing to taking full advantage of local message, carries out partial analysis and process to image, often can obtain better strengthening effect than existing method, especially more obvious to the image advantage of content complexity.In addition, due to local fine process, the generation of the distortion phenomenons such as excessive enhancing, halation is effectively avoided.

(2) present invention employs the method for data-driven, all parameters all by automatically obtaining characteristics of image COMPREHENSIVE CALCULATING, need the problem of manual adjustments parameter when avoiding the image procossing to different characteristic.

(3) the partial analysis process of the present invention's employing and the conbined usage of data driven technique, achieve and strengthen process to the height adaptive of different images, while guarantee strengthens effect, substantially increase broad applicability.

Below in conjunction with accompanying drawing, further detailed description is done to the present invention.

Accompanying drawing explanation

Fig. 1 is algorithm schematic diagram of the present invention.

Fig. 2 is the every step process design sketch of the present invention.

Fig. 3 is the Contrast on effect of the present invention and existing method process normal exposure image.

Fig. 4 is the Contrast on effect of the under-exposed image of the present invention and existing method process.

Fig. 5 is the Contrast on effect of the present invention and the over-exposed image of existing method process.

Fig. 6 is the Contrast on effect of the present invention and existing method process reversible-light shooting image.

Fig. 7 is that the present invention mixes the Contrast on effect of the complicated image of many characteristics with existing method process.

Embodiment

A kind of adapting to image contrast enhancement process based on data-driven of the present invention, comprises the following steps:

Step 1, image is transformed into form and aspect, colourity, gray scale color space from three primary colors color space; Formula used is:

v＝max

Wherein, r, g, b are respectively the value of pixel red, green, blue passage, and max is equivalent to the maximum in r, g, b, and min equals the reckling in these values, and h is the angle value of chrominance space, and value set standardization is between 0 to 360 °.

Step 2, in gray channel, adopt the window of M × N size to travel through with the sliding window method that sliding step is 1 pixel image, wherein M, N are respectively line number and the columns of image subblock; Be specially:

Step 2-1, the subimage in window is carried out to the analysis of gray-scale value and contrast, first, obtain subimage grey level histogram, formula used is:

$\mathrm{pdf}\left(l\right)=\frac{n_l}{M\×N},$

Wherein, l (l ∈ [0,255]) is gray level, n _lbe the number of the pixel of l for gray scale, M, N are respectively line number and the columns of image subblock;

Then, its gray average and contrast is calculated according to image grey level histogram:

$\mathrm{avg}=\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}[\frac{l}{l_{\max }}\·\mathrm{pdf}\left(l\right)],$

$\mathrm{con}\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}[{(l-\mathrm{avg})}^2\·\mathrm{pdf}\left(l\right)]$

Wherein, l _maxit is the maximal value of l;

Step 2-2, according to the grey level histogram of step 2-1 neutron image and the contrast of gray channel, set up the parameterized gray probability distribution function of subimage; Be specially:

${\mathrm{pdf}}^{\′}\left(l\right)={\mathrm{pdf}}_{\mathrm{mac}}\·{\left(\frac{\mathrm{pdf}\left(l\right)-{\mathrm{pdf}}_{\min }}{{\mathrm{pdf}}_{\max }-{\mathrm{pdf}}_{\min }}\right)}^{(1+\sqrt{\mathrm{con}})}$

Wherein, pdf _minand pdf _maxfor the minimum value in each element of grey level histogram vector and maximal value;

The subimage parametrization gray probability distribution function that step 2-3, the eigenwert of subimage extracted according to step 2-1 and step 2-2 obtain, comprehensively goes out to be applicable to the enhancing function of this subimage, is specially:

$\mathrm{cf}\left(l\right)=l_{\max }{\left(\frac{l}{l_{\max }}\right)}^{1+\mathrm{cv}\·\sqrt{\mathrm{gn}\left(l\right)}}$

Wherein, $\mathrm{gn}\left(l\right)=\underset{l=1}{\overset{l}{\mathrm{\Σ}}}\frac{{\mathrm{pdf}}^{\′}\left(l\right)}{\mathrm{sum}\left({\mathrm{pdf}}^{\′}\right)},\mathrm{cv}=\left\{\begin{array}{cc}1,& (\mathrm{avg}\≥180)\\ -1,& \left(\mathrm{else}\right)\end{array}\right.;$ Wherein, sum () is the summation operation to vectorial all elements;

Step 2-4, the gray-scale value of pixel each in subimage carried out to enhancing process, formula used is:

gray'(i,j)＝cf[gary(i,j)]

Wherein, gray (i, j) is original gray value, gray'(i, j) be the gray-scale value after enhancing, (i, j) is the coordinate of current pixel point in view picture figure.

Step 3, the result gray-scale value of sliding window method traversal gray channel to be normalized, the gray channel image after being enhanced; Formula used is:

${\mathrm{gray}}^{\′}(i,j)=\frac{{\mathrm{\Σ}}_{n=1}^{M\×N}G\left(n\right)}{M\×N}$

Wherein, G (n) to be processed the set of the gray-scale value obtained for each pixel by M × N time.

Step 4, the chrominance channel of image to be revised; Specifically carry out revising according to the grey scale change degree before and after strengthening, described strengthen before and after grey scale change degree be:

$\mathrm{dg}=\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}\frac{l}{l_{\max }}[{\mathrm{pdf}}_e\left(l\right)-{\mathrm{pdf}}_o\left(l\right)]$

Wherein, pdf _eand pdf _obe respectively the grey level histogram of image before and after strengthening, revised chrominance representation is:

C (s)=s ^1-dg, wherein s is the chromatic value of each pixel.

Step 5, the gray channel after strengthening to be merged with revised chrominance channel and original form and aspect passage, and be again transformed into rgb color space, the coloured image after being finally enhanced.Formula used is:

$f=\frac{h}{60}-h_i$

p＝v×(1-s)

q＝v×(1-f×s)

t＝v×(1-(1-f)×s)

$(r,g,b)=\begin{array}{c}\left\{\begin{array}{c}(v,t,p),\mathrm{if}{h}_i=0\\ (q,v,p),\mathrm{if}{h}_i=1\\ (p,v,t),\mathrm{if}{h}_i=2\\ (p,q,v),\mathrm{if}{h}_i=3\\ (t,p,v),\mathrm{if}{h}_i=4\\ (v,p,q),\mathrm{if}{h}_i=5\end{array}\right.\end{array}$

Wherein, h is form and aspect, and s is colourity, and v is brightness, and r is red color component value, and g is green component values, and b is blue color component value, h _iby to h divided by 60 business round mould 6 downwards and try to achieve.

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

As shown in Figure 1, implementation step of the present invention is as follows:

Step one, as in Fig. 2, Fig. 2 (a) is an image to be reinforced, entire image is transformed to colourity, form and aspect, gray scale color space (HSV) according to conversion formula from original three primary colors color space (RGB), obtain Fig. 2 (b) chroma channel image, Fig. 2 (c) form and aspect channel image, Fig. 2 (d) gray channel image;

Step 2, use sliding window to travel through to the gray channel of entire image, window sliding step-length is 1 pixel, to the image subblock slided each time in rear window, extracts the characteristic quantity of its mean flow rate avg and contrast con as this region:

$\mathrm{avg}=\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}[\frac{l}{l_{\max }}\·\mathrm{pdf}\left(l\right)]$

$\mathrm{con}\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}[{(l-\mathrm{avg})}^2\·\mathrm{pdf}\left(l\right)]$

Wherein, l (l ∈ [0,255]) is gray level, l _maxbe the maximal value of l, pdf is the normalization histogram of sub-block pixel grey scale.What con characterized is the contrast of original image, and for the image that con is less, the gain strengthening operation just should be larger.

Step 3, in order to make the contrast of image and final enhancing function gain parameter set up organically contact, introduces a kind of parameterized gray-scale value probability distribution function relevant to contrast:

${\mathrm{pdf}}^{\′}\left(l\right)={\mathrm{pdf}}_{\mathrm{mac}}\·{\left(\frac{\mathrm{pdf}\left(l\right)-{\mathrm{pdf}}_{\min }}{{\mathrm{pdf}}_{\max }-{\mathrm{pdf}}_{\min }}\right)}^{(1+\sqrt{\mathrm{con}})}$

Pdf in formula _maxand pdf _minbe maximal value and the minimum value of each element in one-dimensional vector pdf respectively, pdf is normalization discrete grey value probability distribution function:

$\mathrm{pdf}\left(l\right)=\frac{n_l}{M\×N}$

In formula wherein, l (l ∈ [0,255]) is gray level, n _lbe the number of the pixel of l for gray scale, M, N are respectively line number and the columns of image subblock.

Step 4, sets up parameterized enhancing function, the enhancing parameter that this function has cv and gn two adjustable, controls concavity and the gain size of enhancing function respectively, by regulating these parameters, just can obtain different enhancing function curves.With enhancing parameter characteristic of correspondence amount in effective extraction image, and give rational numerical value to adjustable parameter according to this, algorithm has just possessed adaptive ability.Cf is as follows for this enhancing function:

$\mathrm{cf}\left(l\right)=l_{\max }{\left(\frac{l}{l_{\max }}\right)}^{1+\mathrm{cv}\·\sqrt{\mathrm{gn}\left(l\right)}}$

When the image that process brightness is on the low side, usually need to process it by convex function; And when processing bright image, the process of application concave function.The image average avg above calculated is exactly the sign of image subblock average luminous characteristics, therefore can according to this value determination enhancing function concavity controling parameters value:

$\mathrm{cv}=\left\{\begin{array}{cc}1,& (\mathrm{avg}\≥180)\\ -1,& \left(\mathrm{else}\right)\end{array}\right.$

The image that contrast is lower needs to use the larger enhancing function of gain to process it, just can obtain desirable enhancing result, and the image that contrast is high, the gain of corresponding enhancing function then should be less.Special gray probability distribution function pdf' defined above provides the mechanism that a kind of gain coefficient associates with picture contrast.By this mechanism, it is as follows that we define gain control parameter:

$\mathrm{gn}\left(l\right)=\underset{l=1}{\overset{l}{\mathrm{\Σ}}}\frac{{\mathrm{pdf}}^{\′}\left(l\right)}{\mathrm{sum}\left({\mathrm{pdf}}^{\′}\right)}$

Wherein, sum () is sum operation.By above method, the relation that we just establish picture material and strengthen between parameter, by this relation mechanism, this algorithm just can regulate the correlation parameter of enhancing function automatically according to the feature of image itself, generate desirable enhancing function curve.By each pixel in the enhancing function process image subblock automatically generated, the result after just can being enhanced:

gray'(i,j)＝cf[gary(i,j)]。

Step 5, image is divided into some sub-blocks, separately the mutual interference that analyzing and processing just can avoid the different qualities region when processing vision-mix is carried out to these sub-blocks, effective extraction characteristic quantity, obtain correcting function more accurately, and process image subblock according to this, finally these sub-blocks are synthesized complete enhancing image, just can obtain desirable enhancing result, as Fig. 2 (e).The concrete steps of algorithm as follows:

(1) select suitable sub-block width M and height N according to picture size, General Requirements picture size is the integral multiple of the size of sub-block, and the size of sub-block is less, better to the enhancing effect of details.

(2) scan whole image from left to right, from the top down, by the method introduced above, each sub-block is analyzed respectively, process and stores processor result, travel through whole image and then terminate scanning.

(3) each pixel p ix (i, j) has been processed M × N time, obtains a gray scale set G _{(i, j)}(M × N), then enhancing result gray'(i, the j of final synthesis) be:

${\mathrm{gray}}^{\′}(i,j)=\frac{{\mathrm{\Σ}}_{n=1}^{M\×N}G\left(n\right)}{M\×N}.$

Step 6, cause the distortion of color to revise the change strengthening front and back entire image intensity profile, we portray the intensity of variation strengthening front and back gray scale: wherein pdf _eand pdf _obe respectively image grey level histogram before and after strengthening, and revise the chrominance channel strengthening rear image according to this characteristic quantity, revised chrominance representation is: C (s)=s ^1-dg, wherein S is the chromatic value of each pixel, and the image obtained is as Fig. 2 (f).

Step 7, by the gray channel after enhancing, revised chrominance channel and original form and aspect passage merge, and are again transformed into rgb color space, just obtain the coloured image after enhancing, as Fig. 2 (g).

Effect of the present invention can be illustrated by following the simulation experiment result:

We select the complicated image exposing normal, under-exposed, over-exposed, backlight and mix multifrequency nature to come extensive adaptability and the treatment effect of testing algorithm.To every width image, use the method for the low bright algorithm for image enhancement (CA-CD) of histogram equalization (HE), the contrast enhancement algorithms (AGCWD) of weight allocation self-adaptation Gamma correction, content-based hyperchannel and invention to process respectively, and respective result is carried out across comparison.Specific experiment result is as follows:

In Fig. 3, Fig. 3 (a) is the picture obtained under a width normal exposure state, and all the other pictures are through the result images after each algorithm process, is mainly used to test when without the need to strengthening process, the Output rusults of various algorithm.Fig. 3 (b) is the result of HE algorithm, and because this algorithm does not have adaptation mechanism, the process that therefore strengthens is blindly, even if picture exposure is normal, HE also can with very high gain suppression image, and distortion is comparatively serious.Due to the characteristic of algorithm itself, make the result after processing often with the blocking effect distortion of highlight regions, the highlight regions of continuous gray scale change originally can become discontinuous; Fig. 3 (c) is the result of AGCWD algorithm, and this is a kind of adaptive algorithm, has certain adaptive ability.But the enhancing function of this algorithm is fixing and imparametrization, and carries out analyzing and processing for whole image, does not consider local feature.Image after enhancing is partially bright, and highlight regions contrast is lost; Fig. 3 (d) is the result of CA-CD algorithm, and this algorithm is also adaptive, and obtains local message by the neighborhood territory pixel gray scale of each pixel of scanning, and calculates enhancing function.As can be seen from the results, detail section has had further lifting, and overall color gray scale keeps good, and effectiveness comparison is desirable; Fig. 2 (e) is the inventive method, because former figure has uniform intensity profile and higher contrast, therefore the gain coefficient of enhancing function is very low, to the lifting small further of details, on sense organ, whole image does not change substantially, illustrate that this algorithm has good hold facility, successful is better than Fig. 3 (b), Fig. 3 (c), Fig. 2 (d), substantially suitable with Fig. 3 (e).

In Fig. 4, the picture obtained under the under-exposed state of Fig. 4 (a) width, all the other are through the result images after each algorithm process, are mainly used to the treatment effect testing various algorithm in the partially dark situation of image.Fig. 4 (b) and Fig. 4 (c) also exists same problem, and dark portion details obtains obvious lifting, but the mean flow rate of highlighted part is also promoted, and resolution declines to some extent; In Fig. 4 (d), the resolution of highlights and dark portion all has small elevation, but gain is less than normal, strengthens poor effect; The general effect of Fig. 4 (e) is still best, and dark portion details promotes obviously, and it is relatively desirable that highlights keeps, and entire gain is moderate.

In Fig. 5, Fig. 5 (a) is the picture obtained under an over-exposed state, and all the other are through the result images after each algorithm process, is mainly used to the treatment effect testing various algorithm under image crosses bright situation.Fig. 5 (b) serious distortion, gray scale uncontinuity is obvious; There is similar problem in Fig. 5 (c) and Fig. 5 (d), be positioned in the middle part of image, the problem declined due to the over-exposed highlight regions resolution caused fails effectively to be solved, and become on the contrary and be more difficult to identification, and overall brightness is further improved; In Fig. 5 (e) figure, the problems referred to above are well solved, and can identification obviously promote.The normal part details of brightness has small elevation, and overall brightness obtains good maintenance.

In Fig. 6, Fig. 6 (a) is the picture obtained under a width backlight state, and all the other are through the result images after each algorithm process, be mainly used to test processes cross bright and cross dark areas mixing picture time various algorithm effect.In Fig. 6 (b), dark portion details promotes comparatively obvious, but while lifting dark portion details, the brightness of highlight area is also promoted simultaneously, and halation area increases, and highlighted part gray scale uncontinuity is serious; Do not occur gray scale uncontinuity in Fig. 6 (c), but there is the problem identical with Fig. 6 (b), highlights is not well processed; Fig. 6 (d) strengthens low effort, and the result after enhancing does not substantially change compared with former figure; Fig. 6 (e) figure is while lifting dark portion details, and highlights have also been obtained good process, and halation area does not expand, and brightness is not enhanced, and whole structure is desirable.

In Fig. 7, Fig. 7 (b) is the complicated picture that a width is mixed with mist, dark portion, highlights, all the other results after each algorithm process, is mainly used to the effect of various algorithm during the complicated picture of test processes.In Fig. 7 (b), dark portion strengthens successful, but the contrast decline that haze causes does not get a promotion, and has occurred the speck of bulk.In Fig. 7 (c), dark portion strengthens successful, but overall brightness is enhanced, and the contrast that mist causes declines and fails to get a promotion, and the resolution of highlights reduces on the contrary; Promote to some extent the details in the contrast decline region that haze causes in Fig. 7 (d), highlighted and low bright area contrast declines on the contrary to some extent.The details of Fig. 7 (e) to dark portion and highlights all has obvious lifting, and the contrast decline that mist causes simultaneously also has process to a certain degree, and whole structure is obvious.

To sum up, the algorithm that the present invention proposes has stronger adaptive ability to picture of different nature, and the whole structure after process is obviously better than other algorithms.

Claims (6)

1., based on an adapting to image contrast enhancement process for data-driven, it is characterized in that, comprise the following steps:

Step 1, image is transformed into form and aspect, colourity, gray scale color space from three primary colors color space;

Step 2, in gray channel, adopt the window of M × N size to travel through with the sliding window method that sliding step is 1 pixel image, wherein M, N are respectively line number and the columns of image subblock;

Step 3, the result gray-scale value of sliding window method traversal gray channel to be normalized, the gray channel image after being enhanced;

Step 4, the chrominance channel of image to be revised;

Step 5, the gray channel after strengthening to be merged with revised chrominance channel and original form and aspect passage, and be again transformed into rgb color space, the coloured image after being finally enhanced.

2. the adapting to image contrast enhancement process based on data-driven according to claim 1, is characterized in that, in step 1, image is transformed into form and aspect from three primary colors color space, colourity, gray scale color space formula used be:

v＝max

Wherein, r, g, b are respectively the value of pixel red, green, blue passage, and max is equivalent to the maximum in r, g, b, and min equals the reckling in these values, and h is the angle value of chrominance space, and value set standardization is between 0 to 360 °.

3. the adapting to image contrast enhancement process based on data-driven according to claim 1, is characterized in that, step 2 adopts the window of M × N size to travel through with the sliding window method that sliding step is 1 pixel image in gray channel, is specially:

Step 2-1, the subimage in window is carried out to the analysis of gray-scale value and contrast, first, obtain subimage grey level histogram, formula used is:

$\mathrm{pdf}\left(l\right)=\frac{n_l}{M\×N},$

Wherein, l (l ∈ [0,255]) is gray level, n _lbe the number of the pixel of l for gray scale, M, N are respectively line number and the columns of image subblock;

Then, its gray average and contrast is calculated according to image grey level histogram:

$\mathrm{avg}=\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}[\frac{l}{l_{\max }}\·\mathrm{pdf}\left(l\right)],$

$\mathrm{con}=\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}[{(l-\mathrm{avg})}^2\·\mathrm{pdf}\left(l\right)]$

Wherein, l _maxit is the maximal value of l;

Step 2-2, according to the grey level histogram of step 2-1 neutron image and the contrast of gray channel, set up the parameterized gray probability distribution function of subimage; Be specially:

${\mathrm{pdf}}^{\′}\left(l\right)={\mathrm{pdf}}_{\max }\·{\left(\frac{\mathrm{pdf}\left(l\right)-{\mathrm{pdf}}_{\min }}{{\mathrm{pdf}}_{\max }-{\mathrm{pdf}}_{\min }}\right)}^{(1+\sqrt{\mathrm{con}})}$

Wherein, pdf _minand pdf _maxfor the minimum value in each element of grey level histogram vector and maximal value;

The subimage parametrization gray probability distribution function that step 2-3, the eigenwert of subimage extracted according to step 2-1 and step 2-2 obtain, comprehensively goes out to be applicable to the enhancing function of this subimage, is specially:

$\mathrm{cf}\left(l\right)=l_{\max }{\left(\frac{l}{l_{\max }}\right)}^{1+\mathrm{cv}\·\sqrt{\mathrm{gn}\left(l\right)}}$

Wherein, $\mathrm{gn}\left(l\right)=\underset{l=1}{\overset{l}{\mathrm{\Σ}}}\frac{{\mathrm{pdf}}^{\′}\left(l\right)}{\mathrm{sum}\left({\mathrm{pdf}}^{\′}\right)},\mathrm{cv}=\left\{\begin{array}{cc}1,& (\mathrm{avg}\≥180)\\ -1,& \left(\mathrm{else}\right)\end{array}\right.;$ Wherein, sum () is the summation operation to vectorial all elements;

Step 2-4, the gray-scale value of pixel each in subimage carried out to enhancing process, formula used is:

gray'(i,j)＝cf[gary(i,j)]

Wherein, gray (i, j) is original gray value, gray'(i, j) be the gray-scale value after enhancing, (i, j) is the coordinate of current pixel point in view picture figure.

4. the adapting to image contrast enhancement process based on data-driven according to claim 1, is characterized in that, the result gray-scale value of step 3 to sliding window method traversal gray channel is normalized formula used and is:

${\mathrm{gray}}^{\′}(i,j)=\frac{{\mathrm{\Σ}}_{n=1}^{M\×N}G\left(n\right)}{M\×N}$

Wherein, G (n) to be processed the set of the gray-scale value obtained for each pixel by M × N time.

5. the adapting to image contrast enhancement process based on data-driven according to claim 1, it is characterized in that, when the chrominance channel of step 4 pair image is revised, specifically carry out revising according to the grey scale change degree before and after strengthening, described strengthen before and after grey scale change degree be:

$\mathrm{dg}=\underset{l=1}{\overset{l_{\max }}{\mathrm{\Σ}}}\frac{l}{l_{\max }}[{\mathrm{pdf}}_e\left(l\right)-{\mathrm{pdf}}_o\left(l\right)]$

Wherein, pdf _eand pdf _obe respectively the grey level histogram of image before and after strengthening, revised chrominance representation is: C (s)=s ^1-dg, wherein s is the chromatic value of each pixel.

6. the adapting to image contrast enhancement process based on data-driven according to claim 1, it is characterized in that, gray channel after enhancing merges with revised chrominance channel and original form and aspect passage by step 5, and is again transformed into rgb color space, and formula used is:

$f=\frac{h}{60}-h_i$

p＝v×(1-s)

q＝v×(1-f×s)

t＝v×(1-(1-f)×s)

$(r,g,b)=\left\{\begin{array}{ccc}(v,t,p),& \mathrm{if}& h_i=0\\ (q,v,p),& \mathrm{if}& h_i=1\\ (p,v,t),& \mathrm{if}& h_i=2\\ (p,q,v),& \mathrm{if}& h_i=3\\ (t,p,v),& \mathrm{if}& h_i=4\\ (v,p,q),& \mathrm{if}& h_i=5\end{array}\right.$

Wherein, h is form and aspect, and s is colourity, and v is brightness, and r is red color component value, and g is green component values, and b is blue color component value, h _iby to h divided by 60 business round mould 6 downwards and try to achieve.