CN105046658A - Low-illumination image processing method and device - Google Patents

Abstract

The invention provides a low-illumination image processing method and device. A noise suppression filter is additionally arranged before operation of contrast enhancement by aiming at a problem of noise amplification existing in an original low-illumination image contrast enhancement technology, and smoothing is performed on the inverse color image of a low-illumination image by adopting a first filtering coefficient and a second filtering coefficient so that image contrast is enhanced and random noise is suppressed simultaneously. Texture and noise level parameters of the image are calculated according to the local block interior characteristics of the image. Weighted averaging is performed on a first smooth image and a second smooth image after smoothing according to the texture and the noise level parameters. A texture image is obtained by performing texture structure extraction on the gradient image of the inverse color image, and the texture image is combined with the weighted image and the weighted image is sharpened so that the effect of enhancing image details can be realized. Therefore, contrast of the low-illumination image can be effectively enhanced, various types of noise can be filtered, the image color and details can be retained and thus a clear and lifelike restored image can be obtained.

Description

A kind of low-light (level) image processing method and device

Technical field

The application relates to digital image processing field, is specifically related to a kind of low-light (level) image processing method and device.

Background technology

People are usage monitoring camera more and more, is used for ensureing the safety of the aspects such as city, traffic, public place, also uses vehicle-mounted camera to improve the security of driving simultaneously.But low light can reduce the performance of this type of camera greatly according to (as night, backlight, indoor etc.) condition, the image making it take and video visibility reduce, and are often difficult to distinguish the information such as key person, thing.The image taken in above-mentioned situation is called low-light (level) image, and low-light (level) image has various noise, more highlights after carrying out image enhaucament, thus reduce perpetual object in image can identification, the strong subjective feeling reducing people.

For traditional image enhancement technique (such as mist elimination haze and low-light (level) strengthen technology), after it processes image, greatly original noise in image be can strengthen, large stretch of color noise (Colornoise) and some brightness noises (Lumanoise) often occurred in the picture.Traditional image enhancement technique can not these noises of filtering effectively.

For traditional noise-reduction method, it can not solve two problems below:

1) color saturation of object is ensured while the large stretch of color noise of effective filtering.

2) while filtering brightness noise, retain the details of object.

Summary of the invention

The application provides a kind of low-light (level) image processing method and device, solves the problem such as noise amplification, loss of detail after low-light (level) image procossing.

According to the first aspect of the application, this application provides a kind of low-light (level) image processing method, comprising:

Input low-light (level) image;

Described low-light (level) Iamge Segmentation is become different texture regions, calculate the standard deviation of pixel gray scale in each texture region and gradient and, and using described standard deviation and gradient and ratio as the texture of image and noise level parameter;

Inverse process is carried out to described low-light (level) image, obtains inverse image;

According to the mean value of the standard deviation of pixel gray scale in each texture region, determine the first filter factor and the second filter factor, adopt the first filter factor and the second filter factor to the smoothing process of inverse image respectively, obtain the first smoothed image and the second smoothed image respectively;

According to described texture and noise level parameter, the first smoothed image and the second smoothed image are weighted on average, obtain weighted image;

Calculate the dark figure of described weighted image, and obtain environment illumination intensity according to described dark figure;

Contrast strengthen coefficient is obtained according to described dark figure and environment illumination intensity;

Calculate the gradient image of described inverse image, texture structure extraction is carried out to described gradient image, obtains texture image;

Described weighted image is added with texture image, obtains sharpening image;

Carry out contrast strengthen according to described environment illumination intensity and contrast strengthen coefficient to described sharpening image, be enhanced image;

Inverse process is carried out to described enhancing image, obtains output image.

According to the second aspect of the application, this application provides a kind of low-light (level) image processing apparatus, comprising:

Load module, for inputting low-light (level) image;

Image segmentation module, for becoming different texture regions by described low-light (level) Iamge Segmentation;

First computing module, for calculate the standard deviation of pixel gray scale in each texture region and gradient and, and using described standard deviation and gradient and ratio as the texture of image and noise level parameter;

First inverse module, for carrying out inverse process to described low-light (level) image, obtains inverse image;

Smothing filtering module, for the mean value according to the standard deviation of pixel gray scale in each texture region, determine the first filter factor and the second filter factor, adopt the first filter factor and the second filter factor to the smoothing process of inverse image respectively, obtain the first smoothed image and the second smoothed image respectively;

Weighting block, for according to described texture and noise level parameter, is weighted on average the first smoothed image and the second smoothed image, obtains weighted image;

Second computing module, for calculating the dark figure of described weighted image, and obtains environment illumination intensity according to described dark figure; Contrast strengthen coefficient is obtained according to described dark figure and environment illumination intensity;

3rd computing module, for calculating the gradient image of described inverse image, carrying out texture structure extraction to described gradient image, obtaining texture image;

Sharpening module, for being added with texture image by described weighted image, obtains sharpening image;

Contrast-enhancement module, for carrying out contrast strengthen according to described environment illumination intensity and contrast strengthen coefficient to described sharpening image, be enhanced image;

Second inverse module, for carrying out inverse process to described enhancing image, obtains output image.

In the low-light (level) image processing method that the application provides and device, first the low-light (level) Iamge Segmentation of input is become different texture regions, to obtain texture and the noise level parameter of image.On the one hand, according to the mean value of the standard deviation of pixel gray scale in each texture region, determine the first filter factor and the second filter factor, adopt the first filter factor and the second filter factor to the smoothing process of inverse image of low-light (level) image respectively, obtain the first smoothed image and the second smoothed image; According to texture and noise level parameter, the first smoothed image and the second smoothed image are weighted on average, obtain weighted image; Dark figure according to weighted image obtains environment illumination intensity, and then obtains contrast strengthen coefficient.On the other hand, texture structure extraction is carried out to the gradient image of inverse image, obtains texture image.Afterwards, weighted image is added with texture image, obtains sharpening image; Carry out contrast strengthen according to contrast strengthen coefficient to sharpening image again, be enhanced image.Finally inverse process is carried out to enhancing image, obtain output image.Therefore, the low-light (level) image processing method that the application provides and device, can strengthen the contrast of low-light (level) image effectively, the various noise of filtering, retains image color and details simultaneously, obtain clear restored image true to nature.

Accompanying drawing explanation

Fig. 1 is the structural representation of low-light (level) image processing apparatus in a kind of embodiment of the application;

Fig. 2 is the schematic flow sheet of low-light (level) image processing method in a kind of embodiment of the application;

Fig. 3 for the low-light (level) image processing method adopting prior art and the present embodiment respectively and provide same input picture is processed after contrast schematic diagram.

Embodiment

The low-light (level) image processing method that the embodiment of the present application provides and device, can be applicable to video monitoring system, image processing software etc., effectively can carry out the process of mist elimination haze and low-light (level) enhancing process, can filtering color noise and brightness noise on image noise reduction, also retain color and the details of image most possibly.

By reference to the accompanying drawings the application is described in further detail below by embodiment.

Please refer to Fig. 1, present embodiments provide a kind of low-light (level) image processing method and device.Low-light (level) image processing apparatus comprises load module 101, image segmentation module 102, first computing module 103, first inverse module 104, smothing filtering module 105, weighting block 106, second computing module 107, the 3rd computing module 108, sharpening module 109, contrast-enhancement module 110 and the second inverse module 111.

Be described this device below in conjunction with low-light (level) image processing method, please refer to Fig. 2, low-light (level) image processing method comprises step below:

Step 1.1: load module 101 inputs low-light (level) image I.

Step 1.2: low-light (level) image I is divided into different texture regions by image segmentation module 102.Preferably, image segmentation module 102 adopts super-pixel (superpixel) to split and low-light (level) image I is divided into different texture regions.

Step 1.3: the first computing module 103 calculates the standard deviation sigma of pixel gray scale in each texture region and gradient and s, and using the ratio of standard deviation sigma and gradient and s as the texture of image and noise level parameter alpha, i.e. α=σ/s.

Step 1.4: each Color Channel of the first inverse module 104 pairs low-light (level) image I carries out inverse process, obtains inverse image R.First inverse module 104 carries out inverse process according to formula R=255-I to low-light (level) image I.First inverse is carried out to low-light (level) image I, then carrying out subsequent treatment, can low brightness pixel in image be converted to high luminance pixel, thus be conducive to the contrast enhancement processing to low-light (level) region.

Step 1.5: the 3rd computing module 108 utilizes three Color Channels of differentiating operator and R to carry out convolution respectively, obtains the gradient image R of R _d.

Step 1.6: the 3rd computing module 108 selects sizeable filter factor, to R _dcarry out texture structure extraction, obtain R _dthe main texture image R of not Noise _ds.Wherein, filter factor can select empirical value.

Step 1.7: smothing filtering module 105 adopts the first filter factor and the second filter factor to the smoothing process of inverse image respectively, in the present embodiment, utilizes three-dimensional bits to mate (BM3D) wave filter to the smoothing process of inverse image.Use BM3D image filter, while removal picture noise, more image texture details can be retained as far as possible.Preferably, the first filter factor is greater than the mean value of the standard deviation sigma of pixel gray scale in each texture region, and the second filter factor is less than the mean value of the standard deviation sigma of pixel gray scale in each texture region.

Step 1.8: in the present embodiment, smothing filtering module 105 adopts 2 times of the mean value of σ as the first filter factor, obtains the first smoothed image that texture is more level and smooth

Step 1.9: smothing filtering module 105 adopts 1/2 of the mean value of σ as the second filter factor, obtains the second smoothed image that texture is comparatively outstanding

In other embodiments, the first filter factor and the second filter factor can be selected according to the actual requirements, and smothing filtering module 105 also can select other wave filters to the smoothing process of inverse image.

Step 1.10: weighting block 106 is according to texture and noise level parameter alpha, and the first smoothed image obtain filtering and noise reduction and the second smoothed image are weighted on average, obtain weighted image.In the present embodiment, formula is below adopted to obtain weighted image R _s:

$R_s=\mathrm{\α}\·R_s^{fine}+(1-\mathrm{\α})\·R_s^{coarse}$

Step 1.7-step 1.10 be equivalent to carry out contrast strengthen operation front end add noise inhibiting wave filter, solve the noise scale-up problem existed in original low-light (level) picture superposition technology.

Step 1.11: weighted image is added with texture image by sharpening module 109, obtains sharpening image.In the present embodiment, formula is below adopted to obtain the sharpening image R of details enhancing _sharp:

R _sharp＝R _s+α*R _ds

Good sharpen effect can be obtained by weighted sum, avoid edge too to strengthen simultaneously.

In step 1.6,1.7,1.11, remove the noise in gradient image, retain key structure information wherein, and by this structural information, sharpening is carried out to image, play the effect strengthening image detail.

Step 1.12: the second computing module 107 calculates the dark figure R of weighted image _dark, dark figure refers to: the gray level image that Color Channel that in image, on each pixel, in three Color Channels, gray-scale value is minimum is formed.

In the present embodiment, adopt formulae discovery dark figure below:

$R_{dark}\left(x\right)=\underset{y\∈\mathrm{\Ω}\left(x\right)}{\min }\left(\underset{c\∈\left\{r,g,b\right\}}{\min }{R}^c\right(y))$

Wherein, x and y represents the position of pixel, and Ω (x) is the neighborhood centered by pixel x, and c represents different color channels.Concrete, Ω (x) be centered by pixel x, the size neighborhood that is 3*3.

Step 1.13: the second computing module 107 obtains environment illumination intensity A according to dark figure.In the present embodiment, by R _darkeach pixel interior sorts from big to small according to gray-scale value, find out the pixel coming front 0.2%, calculate the mean value of these pixels three Color Channel gray scales in weighted image, find out the pixel that mean value is maximum, using the pixel value (gray-scale values of three Color Channels) of this pixel as the estimated value to environment illumination intensity A.

Step 1.14: the second computing module 107 obtains contrast strengthen coefficient t according to dark figure and environment illumination intensity.In the present embodiment, adopt computing formula below:

$t\left(x\right)=1-\mathrm{\ω}*\underset{y\∈\mathrm{\Ω}\left(x\right)}{min}\left(\underset{c\∈\left\{r,g,b\right\}}{min}\frac{R^c\left(y\right)}{A^c}\right)$

Wherein, x and y represents the position of pixel, t (x) for contrast strengthen coefficient, Ω (x) be the neighborhood centered by pixel x, c represents different color channels, and ω is weight correction factor, R ^c(y) for the brightness value of y pixel in c passage, A be environment illumination intensity.Concrete, Ω (x) be centered by pixel x, the size neighborhood that is 3*3.

In order to avoid occurring that contrast excessively strengthens and strengthen not enough problem, in the present embodiment, the gray-scale value (i.e. the brightness of pixel) according to RGB tri-passages of pixel carrys out adjustment factor ω adaptively, and its regulative mode adopts formula below:

$\mathrm{\ω}\left(x\right)={\left(1-{10}^{-\sqrt{\frac{{\mathrm{\Σ}}_{c\∈\left\{r,g,b\right\}}\left(255-I_c\left(x\right)\right)}{3}}}\right)}^2$

Wherein, ω (x) is the weight correction factor of an xth pixel, I _cx () is the gray-scale value of an xth pixel in c passage.

In other embodiments, ω also can get a fixed value, such as, gets 0.85.

Usually, the t (x) that direct utilization calculates above carries out low-light (level) image enhaucament, understand in the extremely low area failures of illumination, cause and strengthen not enough problem, so according to after formula obtains contrast strengthen coefficient t (x) above, also need simply to revise it, be specially: the enhancing coefficient being less than preset value is reduced further.In the present embodiment, correction formula is as follows:

$t\left(x\right)=\left\{\begin{array}{cc}2t^2\left(x\right),& 0<t\left(x\right)<0.5\\ t\left(x\right),& 0.5<t\left(x\right)<1\end{array}\right.$

In other embodiments, also other correcting modes can be adopted.

Step 1.15: contrast-enhancement module 110 environmentally intensity of illumination and contrast strengthen coefficient carries out contrast strengthen to sharpening image, and be enhanced image, namely carries out mist elimination process to sharpening image, reverts to and strengthens image R clearly _clear, in the present embodiment, below employing, recover formula:

$R_{clear}=\frac{R_{sharp}-A}{t}+A$

Step 1.16: the second inverse module 111, for carrying out inverse process to each Color Channel strengthening image, obtains output image J.Second inverse module 111 is according to formula J=255-R _clearto enhancing image R _clearcarry out inverse process.

Please refer to Fig. 3, Fig. 3 (a) is the low-light (level) image of input, the output image that Fig. 3 (b) obtains for adopting classic method directly to carry out contrast strengthen, the output image that the low-light (level) image processing method that Fig. 3 (c) provides for employing the present embodiment obtains.Can analyze from Fig. 3, the output image that the low-light (level) image processing method adopting embodiment to provide obtains has less noise, and can retain color and the details of image.

It will be appreciated by those skilled in the art that, in above-mentioned embodiment, all or part of step of various method can be carried out instruction related hardware by program and completes, this program can be stored in a computer-readable recording medium, and storage medium can comprise: ROM (read-only memory), random access memory, disk or CD etc.

Above content is the further description done the application in conjunction with concrete embodiment, can not assert that the concrete enforcement of the application is confined to these explanations.For the application person of an ordinary skill in the technical field, under the prerequisite not departing from the present application design, some simple deduction or replace can also be made.

Claims (13)

1. a low-light (level) image processing method, is characterized in that, comprising:

Input low-light (level) image;

Described low-light (level) Iamge Segmentation is become different texture regions, calculate the standard deviation of pixel gray scale in each texture region and gradient and, and using described standard deviation and gradient and ratio as the texture of image and noise level parameter;

Inverse process is carried out to described low-light (level) image, obtains inverse image;

According to the mean value of the standard deviation of pixel gray scale in each texture region, determine the first filter factor and the second filter factor, adopt the first filter factor and the second filter factor to the smoothing process of inverse image respectively, obtain the first smoothed image and the second smoothed image respectively;

According to described texture and noise level parameter, the first smoothed image and the second smoothed image are weighted on average, obtain weighted image;

Calculate the dark figure of described weighted image, and obtain environment illumination intensity according to described dark figure;

Contrast strengthen coefficient is obtained according to described dark figure and environment illumination intensity;

Calculate the gradient image of described inverse image, texture structure extraction is carried out to described gradient image, obtains texture image;

Described weighted image is added with texture image, obtains sharpening image;

Carry out contrast strengthen according to described environment illumination intensity and contrast strengthen coefficient to described sharpening image, be enhanced image;

Inverse process is carried out to described enhancing image, obtains output image.

2. the method for claim 1, is characterized in that, adopts super-pixel segmentation that described low-light (level) Iamge Segmentation is become different texture regions.

3. the method for claim 1, is characterized in that, adopts the first filter factor and the second filter factor respectively, utilizes three-dimensional bits matched filter to the smoothing process of inverse image, obtains the first smoothed image and the second smoothed image respectively; Further, described first filter factor is greater than the mean value of the standard deviation of pixel gray scale in each texture region, and described second filter factor is less than the mean value of the standard deviation of pixel gray scale in each texture region.

4. the method as described in any one of claim 1-3, is characterized in that, also comprises the step revised described contrast strengthen coefficient, is specially: reduce further the enhancing coefficient being less than preset value.

5. a low-light (level) image processing apparatus, is characterized in that, comprising:

Load module, for inputting low-light (level) image;

Image segmentation module, for becoming different texture regions by described low-light (level) Iamge Segmentation;

First computing module, for calculate the standard deviation of pixel gray scale in each texture region and gradient and, and using described standard deviation and gradient and ratio as the texture of image and noise level parameter;

First inverse module, for carrying out inverse process to described low-light (level) image, obtains inverse image;

Smothing filtering module, for the mean value according to the standard deviation of pixel gray scale in each texture region, determine the first filter factor and the second filter factor, adopt the first filter factor and the second filter factor to the smoothing process of inverse image respectively, obtain the first smoothed image and the second smoothed image respectively;

Weighting block, for according to described texture and noise level parameter, is weighted on average the first smoothed image and the second smoothed image, obtains weighted image;

Second computing module, for calculating the dark figure of described weighted image, and obtains environment illumination intensity according to described dark figure; Contrast strengthen coefficient is obtained according to described dark figure and environment illumination intensity;

3rd computing module, for calculating the gradient image of described inverse image, carrying out texture structure extraction to described gradient image, obtaining texture image;

Sharpening module, for being added with texture image by described weighted image, obtains sharpening image;

Contrast-enhancement module, for carrying out contrast strengthen according to described environment illumination intensity and contrast strengthen coefficient to described sharpening image, be enhanced image;

Second inverse module, for carrying out inverse process to described enhancing image, obtains output image.

6. device as claimed in claim 5, is characterized in that, described low-light (level) Iamge Segmentation is become different texture regions for adopting super-pixel to split by image segmentation module.

7. device as claimed in claim 5, it is characterized in that, smothing filtering module is used for adopting the first filter factor and the second filter factor respectively, utilizes three-dimensional bits matched filter to the smoothing process of inverse image, obtains the first smoothed image and the second smoothed image respectively; Further, described first filter factor is greater than the mean value of the standard deviation of pixel gray scale in each texture region, and described second filter factor is less than the mean value of the standard deviation of pixel gray scale in each texture region.

8. device as claimed in claim 7, is characterized in that, weighting block is used for according to described texture and noise level parameter, is weighted on average, when obtaining weighted image, adopts formula below to the first smoothed image and the second smoothed image:

$R_S-a\&CenterDot;R_S^{fine}+(1-\mathrm{\&alpha;})\&CenterDot;R_S^{coarse}$

Wherein, R _sfor weighted image, be the first smoothed image, be the second smoothed image, α is texture and noise level parameter.

9. device as claimed in claim 8, is characterized in that, sharpening module is used for described weighted image to be added with texture image, when obtaining sharpening image, adopts formula below:

R _sharp＝R _s+α*R _ds

Wherein, R _sharpfor sharpening image, R _dsfor texture image.

10. device as claimed in claim 5, is characterized in that, when the second computing module is used for obtaining contrast strengthen coefficient according to dark figure and environment illumination intensity, adopts formula below:

$t\left(x\right)=1-\mathrm{\&omega;}*\underset{y\&Element;\mathrm{\&Omega;}\left(x\right)}{min}\left(\underset{c\&Element;\{r,g,b\}}{\min }\frac{R^c\left(y\right)}{A^c}\right)$

Wherein, x and y represents the position of pixel, t (x) for contrast strengthen coefficient, Ω (x) be the neighborhood centered by pixel x, c represents different color channels, and ω is weight correction factor, R ^c(y) for the brightness value of y pixel in c passage, A be environment illumination intensity.

11. devices as claimed in claim 10, is characterized in that, the second computing module also for adopt below formula according to pixel weight correction factor described in the brightness value Automatic adjusument of RGB tri-passages:

$\mathrm{\&omega;}\left(x\right)={(1-10\sqrt[-]{\frac{{\mathrm{\&Sigma;}}_{c\&Element;(r,g,b)}(255-I_c(x))}{3}})}^2$

Wherein, ω (x) is the weight correction factor of an xth pixel, I _cx () is the brightness value of an xth pixel in c passage.

12. devices as described in any one of claim 5-11, it is characterized in that, the second computing module also for revising described contrast strengthen coefficient, is specially: the second computing module is used for reducing further the enhancing coefficient being less than preset value.

13. devices as claimed in claim 12, is characterized in that, when the second computing module is used for revising described contrast strengthen coefficient, adopt formula below:

$t\left(x\right)=\left\{\begin{array}{cc}2t^2\left(x\right),& 0<t\left(x\right)<0.5\\ t\left(x\right),& 0.5<t\left(x\right)<1\end{array}\right.$

Wherein, t (x) is contrast strengthen coefficient.