CN103606134A - Enhancing method of low-light video images - Google Patents

Abstract

The invention relates to an enhancing method of low-light video images. The method comprises the following steps that (1) the optimal scale parameter of a Gaussian function in the single-scale Rctinex; (2) high-light images serve as reference images of color migration; (3) the images having whole rich colors and local outstanding details are obtained. Compared with the prior art, the enhancing method has the advantages of being good in image enhancing effect, obtaining rich image color information and the like.

Description

A kind of low light level is according to the Enhancement Method of video image

Technical field

The present invention relates to a kind of image processing techniques, especially relate to a kind of low light level according to the Enhancement Method of video image.

Background technology

Image (Image) is human transmit message's important media, and in the information that human intelligence is processed, visual communication information accounts for the mankind's the more than 75% of total amount of receiving information.Image is a kind of important means of mankind's obtaining information, the perception world and then reforming world.

Power grid construction work progress relate to project management, civil engineering, very many specialties such as electric, industry cross-operation very obviously, severe, the construction safety risk of construction environment and potential safety hazard more, these are all the difficult problems existing in power grid construction work progress.With the spirit of country " three collection five large " and the demand of the information-based management and control of strengthening engineering construction key node and the safety and stability of project management transition period form contrast be, current work progress, the management and control of working-yard are indifferent, lack the technological means of security monitoring.For this reason, the present invention be take Urban Underground construction of transformer substation as point of penetration, and in the actual conditions of power construction process, construction lighting condition in early stage in underground substation is undesirable, and the video image gray scale obtaining is very low, and visual effect is undesirable.In order to obtain image clearly, by the improvement to Retinex algorithm, realize figure image intensifying.

The development of video enhancement techniques follows hard on the development of video technique to be carried out, and because video is set of number image sequence, therefore the many digital video images that can be applied to of existing algorithm for image enhancement/software strengthen.But digital video strengthens algorithm, also there is own feature, as require the processing of high real-time, frame of video to need continuity etc.

Traditional video enhancement method does not utilize high brightness information to add in enhancing conventionally, only for video itself, adopts enhancing algorithm to process.Video enhancement techniques is summarized as two large classes: airspace enhancement facture (Spatial-based Domain Enhancement) and frequency domain strengthen facture (Frequency-based Domain Enhancement).Airspace enhancement method is normally for pixel operation.Mostly the algorithm based on airspace enhancement belongs to the method for direct augmented video itself, comprises greyscale transformation, histogram transformation, filter process, fuzzy logic enhancing, based on genetic algorithm optimization etc.It is that image is carried out to certain conversion that frequency domain strengthens, and in transform domain, the coefficient after conversion is carried out to computing, and then contravariant changes to original spatial domain, and the image being enhanced, is a kind of indirect disposal route.Conventional image conversion has: Fourier transform, wavelet transformation, discrete cosine transform, Walsh transform and Hotelling transform etc.What wherein application was the most ripe is Fourier transform, it take the Fourier transform of revising image is basis, image is transformed in frequency field, again the frequency field of image is carried out to filtering processing, spatial domain is changed in last contravariant, obtains the image after strengthening, but when video is processed in real time, the complicated real-time of computing is poor, can not meet the demands.

Summary of the invention

Object of the present invention is exactly to provide a kind of figure image intensifying Enhancement Method that the low light level effective, that obtain rich image color information shines video image in order to overcome the defect of above-mentioned prior art existence.

Object of the present invention can be achieved through the following technical solutions:

The low light level according to an Enhancement Method for video image, is characterized in that, comprises the following steps:

1) ask the optimal scale parameter of Gaussian function in single scale Retinex;

2) reference picture using intense light irradiation image as color transfer;

3) obtaining overall color enriches and the outstanding image of local detail.

Described step 1) ask the optimal scale parameter of Gaussian function in single scale Retinex to be specially:

Gaussian convolution function can better obtain luminance picture from original image, and Gaussian convolution function is

$G(x,y)=\mathrm{\λexp}\left(\frac{-(x^2+y^2)}{c^2}\right)---\left(1\right)$

In formula, c is scale parameter; λ is normalized factor, makes

∫∫G(x，y)dxdy＝1 (2)

Gaussian convolution function only has a variable element, and scale parameter c has determined the treatment effect of single scale Retinex, and hour, dynamic range compression is larger for c, and image detail is outstanding, but the loss of overall illumination, image has cross-color phenomenon; When c is larger, integral image is effective, and color fidelity is good, but dynamic range compression is less, and local detail is unintelligible;

If original image is I (x, y), reflected image is R (x, y), and luminance picture is L (x, y), according to formula (1), supposes that luminance picture is level and smooth, and single scale Retinex algorithm can be expressed as in log-domain

$\log R(x,y)=\log \frac{I(x,y)}{L(x,y)}=\log I(x,y)-\log L(x,y)=\log I(x,y)-\log (I(x,y)*G(x,y))---\left(3\right)$

In formula, G (x, y) is low pass convolution function, and luminance picture is

L(x，y)=I(x，y)*G(x，y) (4)

Information entropy is the tolerance to the contained quantity of information of image, and information entropy is larger, and the complexity of presentation video texture is higher, and the contained information of image is more, and information entropy formula is

$\mathrm{entropy}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{p}_i\log p_i---\left(5\right)$

Wherein, p _ithe frequency that i gray level of presentation video occurs, when all gray level probabilities of occurrence equate, the information entropy of image is maximum; When image is monochrome, the information entropy of image is zero;

Average is the average gray value of pixel, reflection image whole chiaroscuro effect, average is less, image is darker; Gray standard deviation has reflected the discrete situation of relative gray average, and standard deviation is larger, and intensity profile is overstepping the bounds of propriety loose, and picture quality is better; One width size is the image of mxn, and its brightness average and standard deviation formula are

$\mathrm{average}=\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}g(i,j)}{m\×n}---\left(6\right)$

$\mathrm{std}2=\sqrt{\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(g(i,j)-\mathrm{average})}^2}{m\×n}}---\left(7\right)$

In formula, g (i, j) is the pixel value of coordinate (i, j);

Through contrast, entropy is maximum, and average and the standard deviation of image are placed in the middle, and figure image intensifying effect is best.

Described step 2) reference picture using intense light irradiation image as color transfer is specially:

At l α β color space, the average as each Color Channel and standard deviation are added up, the integral color information of average represent images wherein, standard deviation is the minutia of represent images;

Preserve average and the standard deviation of color transfer image reference image;

In order to meet the requirement of video image real-time, color transfer reference picture is from Same Scene intense light irradiation on daytime image, and this image data after pre-service are directly called after preserving.

Described step 3) obtaining the abundant and outstanding image of local detail of overall color is specially:

Call the value of optimal scale parameter to the Gaussian function of single scale Retincx, the image after being enhanced by single scale Retinex;

Calculate average and the standard deviation of the image after strengthening;

Call average and the standard deviation of color transfer image reference image;

Average and the standard deviation of color transfer image reference image are dwindled, and guarantee that after strengthening, the color information of image is undistorted;

After the Global Information of reference picture is delivered to and strengthens, in image, after strengthening, the average of image adds the average of reference picture respective channel, obtains composograph.

Compared with prior art, the present invention has the following advantages:

1, by asking for the optimal scale parameter of Gaussian function in single scale Retinex (SSR), now image information entropy is maximum, and figure image intensifying effect is best.

2, the image after being enhanced by single scale Retinex (SSR), although image local detail is outstanding, but overall color is single, for obtaining rich image color information, image after color reference image and SSR enhancing is strengthened by Reinhard color transfer.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of Gaussian function optimal scale parametric solution method that the present invention is based on the single scale Retinex of information entropy;

Fig. 2 is the process flow diagram that the present invention is based on the color information method for solving of Reinhard color transfer;

Fig. 3 is the process flow diagram that the present invention is based on the color information method for solving of Reinhard color transfer.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.

Embodiment

As Figure 1-3, a kind of low light level according to the Enhancement Method of video image, is characterized in that, comprises the following steps:

1) ask the optimal scale parameter of Gaussian function in single scale Retinex;

2) reference picture using intense light irradiation image as color transfer;

3) obtaining overall color enriches and the outstanding image of local detail.

Described step 1) ask the optimal scale parameter of Gaussian function in single scale Retinex to be specially:

Gaussian convolution function can better obtain luminance picture from original image, and Gaussian convolution function is

$G(x,y)=\mathrm{\λexp}\left(\frac{-(x^2+y^2)}{c^2}\right)$

In formula, c is scale parameter; λ is normalized factor, makes

∫∫G(x，y)dxdy=1

Gaussian convolution function only has a variable element, and scale parameter c has determined the treatment effect of single scale Retinex, and hour, dynamic range compression is larger for c, and image detail is outstanding, but the loss of overall illumination, image has cross-color phenomenon; When c is larger, integral image is effective, and color fidelity is good, but dynamic range compression is less, and local detail is unintelligible;

If original image is I (x, y), reflected image is R (x, y), and luminance picture is L (x, y), according to formula (1), supposes that luminance picture is level and smooth, and single scale Retinex algorithm can be expressed as in log-domain

$\log R(x,y)=\log \frac{I(x,y)}{L(x,y)}=\log I(x,y)-\log L(x,y)=\log I(x,y)-\log (I(x,y)*G(x,y))$

In formula, G (x, y) is low pass convolution function, and luminance picture is

L(x，y)=I(x，y)*G(x，y)

Information entropy is the tolerance to the contained quantity of information of image, and information entropy is larger, and the complexity of presentation video texture is higher, and the contained information of image is more, and information entropy formula is

$\mathrm{entropy}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{p}_i\log p_i$

Wherein, p _ithe frequency that i gray level of presentation video occurs, when all gray level probabilities of occurrence equate, the information entropy of image is maximum; When image is monochrome, the information entropy of image is zero;

Average is the average gray value of pixel, reflection image whole chiaroscuro effect, average is less, image is darker; Gray standard deviation has reflected the discrete situation of relative gray average, and standard deviation is larger, and intensity profile is overstepping the bounds of propriety loose, and picture quality is better; One width size is the image of mxn, and its brightness average and standard deviation formula are

$\mathrm{average}=\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}g(i,j)}{m\×n}$

$\mathrm{std}2=\sqrt{\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(g(i,j)-\mathrm{average})}^2}{m\×n}}$

In formula, g (i, j) is the pixel value of coordinate (i, j);

Through contrast, entropy is maximum, and average and the standard deviation of image are placed in the middle, and figure image intensifying effect is best.

Described step 2) reference picture using intense light irradiation image as color transfer is specially:

At l α β color space, the average as each Color Channel and standard deviation are added up, the integral color information of average represent images wherein, standard deviation is the minutia of represent images;

Preserve average and the standard deviation of color transfer image reference image;

In order to meet the requirement of video image real-time, color transfer reference picture is from Same Scene intense light irradiation on daytime image, and this image data after pre-service are directly called after preserving.

Described step 3) obtaining the abundant and outstanding image of local detail of overall color is specially:

Call the value of optimal scale parameter to the Gaussian function of single scale Retinex, the image after being enhanced by single scale Retinex;

Calculate average and the standard deviation of the image after strengthening;

Call average and the standard deviation of color transfer image reference image;

Average and the standard deviation of color transfer image reference image are dwindled, and guarantee that after strengthening, the color information of image is undistorted;

After the Global Information of reference picture is delivered to and strengthens, in image, after strengthening, the average of image adds the average of reference picture respective channel, obtains composograph.

Claims (4)

1. the low light level, according to an Enhancement Method for video image, is characterized in that, comprises the following steps:

1) ask the optimal scale parameter of Gaussian function in single scale Retinex;

2) reference picture using intense light irradiation image as color transfer;

3) obtaining overall color enriches and the outstanding image of local detail.

2. the low light level according to claim 1, according to the Enhancement Method of video image, is characterized in that described step 1) ask the optimal scale parameter of Gaussian function in single scale Retinex to be specially:

Gaussian convolution function can better obtain luminance picture from original image, and Gaussian convolution function is

$G(x,y)=\mathrm{\λexp}\left(\frac{-(x^2+y^2)}{c^2}\right)---\left(1\right)$

In formula, c is scale parameter; λ is normalized factor, makes

∫∫G(x，y)dxdy=1 (2)

Gaussian convolution function only has a variable element, and scale parameter c has determined the treatment effect of single scale Retinex, and hour, dynamic range compression is larger for c, and image detail is outstanding, but the loss of overall illumination, image has cross-color phenomenon; When c is larger, integral image is effective, and color fidelity is good, but dynamic range compression is less, and local detail is unintelligible;

If original image is I (x, y), reflected image is R (x, y), and luminance picture is L (x, y), according to formula (1), supposes that luminance picture is level and smooth, and single scale Retinex algorithm is expressed as in log-domain

$\log R(x,y)=\log \frac{I(x,y)}{L(x,y)}=\log I(x,y)-\log L(x,y)=\log I(x,y)-\log (I(x,y)*G(x,y))---\left(3\right)$

In formula, G (x, y) is low pass convolution function, and luminance picture is

L(x，y)=I(x，y)*G(x，y) (4)

Information entropy is the tolerance to the contained quantity of information of image, and information entropy is larger, and the complexity of presentation video texture is higher, and the contained information of image is more, and information entropy formula is

$\mathrm{entropy}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{p}_i\log p_i---\left(5\right)$

Wherein, p _ithe frequency that i gray level of presentation video occurs, when all gray level probabilities of occurrence equate, the information entropy of image is maximum; When image is monochrome, the information entropy of image is zero;

Average is the average gray value of pixel, reflection image whole chiaroscuro effect, average is less, image is darker; Gray standard deviation has reflected the discrete situation of relative gray average, and standard deviation is larger, and intensity profile is overstepping the bounds of propriety loose, and picture quality is better; One width size is the image of mxn, and its brightness average and standard deviation formula are

$\mathrm{average}=\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}g(i,j)}{m\×n}---\left(6\right)$

$\mathrm{std}2=\sqrt{\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(g(i,j)-\mathrm{average})}^2}{m\×n}}---\left(7\right)$

In formula, g (i, j) is the pixel value of coordinate (i, j);

Through contrast, entropy is maximum, and average and the standard deviation of image are placed in the middle, and figure image intensifying effect is best.

3. the low light level according to claim 1, according to the Enhancement Method of video image, is characterized in that described step 2) reference picture using intense light irradiation image as color transfer is specially:

At l α β color space, the average as each Color Channel and standard deviation are added up, the integral color information of average represent images wherein, standard deviation is the minutia of represent images;

Preserve average and the standard deviation of color transfer image reference image;

In order to meet the requirement of video image real-time, color transfer reference picture is from Same Scene intense light irradiation on daytime image, and this image data after pre-service are directly called after preserving.

4. the low light level according to claim 1, according to the Enhancement Method of video image, is characterized in that described step 3) obtain the abundant and outstanding image of local detail of overall color and be specially:

Call the value of optimal scale parameter to the Gaussian function of single scale Retinex, the image after being enhanced by single scale Retinex;

Calculate average and the standard deviation of the image after strengthening;

Call average and the standard deviation of color transfer image reference image;

Average and the standard deviation of color transfer image reference image are dwindled, and guarantee that after strengthening, the color information of image is undistorted;

After the Global Information of reference picture is delivered to and strengthens, in image, after strengthening, the average of image adds the average of reference picture respective channel, obtains composograph.

Images