CN101345891A - Non-reference picture quality appraisement method based on information entropy and contrast - Google Patents

Abstract

The invention discloses a non-reference image quality evaluation method based on information entropy and contrast, comprising the following steps: 1, obtaining an original image; 2, transforming the original image into a gray level image; 3, computing average gray level of the original image and continuous bandwidth of the gray spectrum; 4, determining the search range of the gray initial value Theta and the gray level Delta; 5, performing Zadeh-X transformation to the image according to Theta and Delta values; 6, computing average contrast and information entropy of the transformed image, constructing a non-reference image quality evaluation function by the product of the average contrast and the information entropy, and computing the value thereof; 7, adding 1 to Delta and returning to step 5; 8, computing the maximum of the evaluation function after finishing research as well as corresponding delta and Theta values, performing Zadeh-X transformation to the original image according to the two values and then obtaining the best quality image. The invention provides the criterion of the best image quality, and obtains the best quality image according to the criterion.

Description

Non-reference picture quality appraisement method based on comentropy and contrast

Technical field

The present invention relates to image processing field, specifically, is that a kind of decline type gray scale that has from 0 gray scale that is used to estimate based on comentropy and contrast is composed the non-reference picture method for quality that distributes.

Background technology

The objective evaluation of picture quality is an important very difficult again problem during computer digital image is handled, current a large amount of research and patents all is about the reference picture quality evaluation is arranged, estimate the situation that degrades of image after treatment, non-reference picture quality appraisement (NR-IQA:no reference image quality assessment) then is the most important and the most difficult task that computer digital image is handled, and it has the application of following three aspects: (1) is used for image/video supervisory control system monitoring image quality; (2) be used to adjust image/video treatment system and algorithm; (3) be used for embedded images/processing system for video with optimized algorithm and parameter setting.But present non-reference picture quality appraisement technology is to estimate the improvement of quality behind the image denoising, does not relate to the criterion of optimum picture quality.

Summary of the invention

The object of the present invention is to provide a kind of non-reference picture quality appraisement method, the criterion that obtains optimum picture quality is provided based on comentropy and contrast.

To achieve these goals, technical scheme of the present invention is as follows: a kind of non-reference picture quality appraisement method based on comentropy and contrast, and its key is to carry out as follows:

(1) obtains original image;

Original image is a digital picture, can pass through acquisitions such as camera, video camera, monitor.

(2) original image is converted to original-gray image;

If original image originally as gray level image, does not then need conversion, if original image then needs to be converted into gray level image originally as coloured image, coloured image is converted to gray level image two kinds of methods:

First kind is to adopt normalization weighted sum formula that coloured image is converted to gray level image:

O(x，y)＝R(x，y)×0.3+G(x，y)×0.59+B(x，y)×0.11

In the formula, R (x, y), G (x, y), (x y) is respectively pixel (x, three kinds of chromatic values of red, green, blue y) to B.(x is y) for changing the gray value of back corresponding points for O.

Second kind is that power conversion formulas such as employing are changed:

O(x，y)＝R(x，y)/3+G(x，y)/3+B(x，y)/3

In the formula, R (x, y), G (x, y), (x y) is respectively pixel (x, three kinds of chromatic values of red, green, blue y) to B.(x is y) for changing the gray value of back corresponding points for O.

(3) (x, y) information are calculated the average gray value G of this original-gray image and gray scale spectrum, and obtain gray scale spectrum continuous bandwidth BW according to this gray scale spectrum to obtain the gray value O of this original-gray image;

Described gray scale spectrum continuous bandwidth is represented one section tonal range of spectral line continuous distribution, begin search from first spectral line of gray scale spectrum, in case do not have spectral line on certain gray scale, promptly no pixel exists on this gray scale, then stop search, the width of this section Continuous Gray Scale spectrum is bandwidth BW.

Described gray scale spectrum is pressed following formula and is obtained:

$T\left(g\right)=\frac{O^{1/m}\left(g\right)}{\underset{g=0}{\overset{255}{\mathrm{\Σ}}}{O}^{1/m}\left(g\right)}\underset{g=0}{\overset{255}{\mathrm{\Σ}}}O\left(g\right)$

In the formula, m is the planarization level, get [1, the ∞) integer between.Correlation theory is seen the Chinese invention patent (patent No.: ZL200610054324.9) that " is used for the gradation of image of bottom layer image mining or the high-resolution detection method of chrominance information ".

(4) determine gray scale initial value Theta and hunting zone; And give tonal range Delta with the starting point assignment of hunting zone;

(5) according to the gray value O of described gray scale initial value Theta, tonal range Delta and original image (x y), carries out conversion with the Zadeh-X transform method to original-gray image, obtain one group of new T of grey scale pixel value as a result (x, y);

Described Zadeh-X transform method adopts following formula to calculate:

$T(x,y)=K\frac{O(x,y)-\mathrm{Theta}}{\mathrm{Delta}}$

Wherein, (x y) is the coordinate of each pixel in the image, O (x, y) scope is [0,255], and T (x, scope y) is [0,255], Theta is an integer, and span is [0,255], Delta is a positive integer, and span is [1,255], K represents contraction-expansion factor, and span is [1,255], be more evenly distributed, generally get K=255 in order to make the gradation of image after the conversion.

When O (x, y)＜during Theta, T (x, y)=0; When O (x, in the time of y)＞(Theta+Delta), T (x, y)=255.Correlation theory see Chinese invention patent " bottom layer image hide and method for digging and adopt the image concealing and the excavating gear of this method " (the patent No.: ZL200610054379.X).

(6) (x y), calculates this average contrast C and comentropy InEn of grey scale pixel value correspondence as a result according to the new T of grey scale pixel value as a result that obtains;

(7) according to average contrast C and comentropy InEn, obtain the value IQAF of evaluation function, described evaluation function is:

IQAF＝C×InEn

(8) Delta is added 1, turn back to step (five) again, terminal point up to the hunting zone;

(9) after search finishes, find out the maximum IQAF and the Delta value of this IQAF correspondence, with the Zadeh-X transform method original-gray image is carried out conversion, (x, y), the image that this value constitutes is best quality image to obtain one group of new T of grey scale pixel value as a result.

In other words, the criterion of optimum picture quality is the maximum of IQAF, the Delta value corresponding according to maximum IQAF value, and (x y), just can obtain the image of best in quality with the Zadeh-X transform method in conjunction with the gray value O of Theta and original image again.

In described step (four), described gray scale initial value Theta is the gray scale spectrum starting point gray value of band continuously, and the starting point of hunting zone is the average gray value G of original-gray image, and terminal point is

The present invention is used to assess the image that has from the decline type gray scale spectrum distribution of 0 gray scale, so Theta generally gets 0 value.And find that in actual search maximum generally exists

The preceding appearance is so the terminal point of hunting zone is made as

Can save search time.

The average gray value G of described gray level image is obtained by following formula:

$\stackrel{\‾}{G}=\frac{1}{M\×N}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\mathrm{Gray}(x,y)$

In the formula, (x is that (M, N are x to pixel, the pixel count of y direction for x, gray scale y) y) to Gray.Because Delta is positive integer, if therefore calculate average gray value G and

Be decimal, then adopt the mode round up, they are adjusted into integer after assignment give Delta.

In described step (six), described average contrast C is obtained by following formula:

$\stackrel{\‾}{C}=\frac{1}{(M-1)\×(N-1)}\underset{y=0}{\overset{N-2}{\mathrm{\Σ}}}\underset{x=0}{\overset{M-2}{\mathrm{\Σ}}}|\mathrm{Gray}(x,y)-\mathrm{Gray}(x+1,y)|$

In the formula, (x is that (M, N are x to pixel, the pixel count of y direction for x, gray scale y) y) to Gray.

In described step (six), described comentropy InEn is obtained by following formula:

$\mathrm{InEn}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}p\left(i\right){\mathrm{Log}}_{2}p\left(i\right)$

In the formula, the probability of the pixel of p (i) expression gray scale i.

Beneficial effect: the present invention proposes a kind of non-reference picture quality appraisement method based on comentropy and contrast, be used to estimate and have the image that distributes from the decline type gray scale spectrum of 0 gray scale, adopting the product of comentropy and contrast is evaluation function, and search for this evaluation function maximum, the image of the corresponding best in quality of this maximum, the criterion of optimum picture quality is provided,, has obtained the image of best in quality according to this evaluation function maximum corresponding parameters.

Description of drawings

Fig. 1 is a workflow schematic diagram of the present invention;

Fig. 2 is an original image;

Fig. 3 is the picture quality comparison diagram, a) is original image shown in Figure 2 wherein, b) is the image of IQAF maximum correspondence, c) is the corresponding image of another IQAF value.

Embodiment

Further the present invention is illustrated below in conjunction with drawings and Examples.

As shown in Figure 1: a kind of non-reference picture quality appraisement method based on comentropy and contrast, carry out as follows:

(1) obtains original image;

Original image is a digital picture, can pass through acquisitions such as camera, video camera, monitor.As shown in Figure 1: the width of cloth gray scale pictures of original image under low lighting condition, taking.

(2) original image is converted to original-gray image;

As shown in Figure 2, original image does not need to change originally as gray level image.If but original image is a coloured image, then needs to be converted into gray level image, coloured image is converted to gray level image two kinds of methods:

First kind is to adopt normalization weighted sum formula to change:

O(x，y)＝R(x，y)×0.3+G(x，y)×0.59+B(x，y)×0.11

In the formula, R (x, y), G (x, y), (x y) is respectively pixel (x, three kinds of chromatic values of red, green, blue y) to B.(x is y) for changing the gray value of back corresponding points for O.

Second kind is that power conversion formulas such as employing are changed:

O(x，y)＝R(x，y)/3+G(x，y)/3+B(x，y)/3

In the formula, R (x, y), G (x, y), (x y) is respectively pixel (x, three kinds of chromatic values of red, green, blue y) to B.(x is y) for changing the gray value of back corresponding points for O.

(3) (x, y) information are calculated the average gray value G of this original-gray image and gray scale spectrum, and obtain gray scale spectrum continuous bandwidth BW according to this gray scale spectrum to obtain the gray value O of this original-gray image;

The average gray value G of described gray level image is obtained by following formula:

$\stackrel{\‾}{G}=\frac{1}{M\×N}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\mathrm{Gray}(x,y)$

In the formula, (x is that (M, N are x to pixel, the pixel count of y direction for x, gray scale y) y) to Gray.The average gray value G of image shown in Fig. 2 is 4.7758, is 5 after rounding up.

Described gray scale spectrum continuous bandwidth is represented one section tonal range of spectral line continuous distribution, begin search from first spectral line of gray scale spectrum, in case do not have spectral line on certain gray scale, promptly no pixel exists on this gray scale, then stop search, the width of this section Continuous Gray Scale is bandwidth BW.

Described gray scale spectrum is pressed following formula and is obtained:

$T\left(g\right)=\frac{O^{1/m}\left(g\right)}{\underset{g=0}{\overset{255}{\mathrm{\Σ}}}{O}^{1/m}\left(g\right)}\underset{g=0}{\overset{255}{\mathrm{\Σ}}}O\left(g\right)$

In the formula, m is the planarization level, get [1, the ∞) integer between.Correlation theory is seen the Chinese invention patent (patent No.: ZL200610054324.9) that " is used for the gradation of image of bottom layer image mining or the high-resolution detection method of chrominance information ".These gray scale spectrum computational methods have the precision of a pixel, calculate the continuous bandwidth BW=66 of gradation of image spectrum among Fig. 2 according to following formula.

(4) determine the hunting zone of gray scale initial value Theta and Delta; Described gray scale initial value Theta is the gray scale spectrum starting point gray value of band continuously, and the starting point of Delta hunting zone is the average gray value G of original-gray image, and terminal point is

The present invention is used to estimate the image that has from the decline type gray scale spectrum distribution of 0 gray scale, so Theta generally gets 0 value.And find that in actual search maximum generally exists

The preceding appearance is so the terminal point of hunting zone is made as

Can save search time. $\frac{1}{5}\mathrm{BW}=66/5=13.2.$ In the present embodiment, the starting point of hunting zone is 5, and terminal point is 13.

(5) according to the gray value O of described gray scale initial value Theta, tonal range Delta and original image (x y), carries out conversion with the Zadeh-X transform method to original-gray image, obtain one group of new T of grey scale pixel value as a result (x, y);

Described Zadeh-X transform method adopts following formula to calculate:

$T(x,y)=K\frac{O(x,y)-\mathrm{Theta}}{\mathrm{Delta}}$

Wherein, (x y) is the coordinate of each pixel in the image, O (x, scope y) is [0,255], T (x, scope y) is [0,255], Theta ∈ [0,255], Delta ∈ [1,255], K represents contraction-expansion factor, span be [, 255], be more evenly distributed, generally get K=255 in order to make the gradation of image after the conversion.

When O (x, y)＜during Theta, T (x, y)=0; When O (x, in the time of y)＞(Theta+Delta), T (x, y)=255.

(6) (x y), calculates this average contrast C and comentropy InEn of grey scale pixel value correspondence as a result according to the new T of grey scale pixel value as a result that obtains;

Described average contrast C is obtained by following formula:

$\stackrel{\‾}{C}=\frac{1}{(M-1)\×(N-1)}\underset{y=0}{\overset{N-2}{\mathrm{\Σ}}}\underset{x=0}{\overset{M-2}{\mathrm{\Σ}}}|\mathrm{Gray}(x,y)-\mathrm{Gray}(x+1,y)|$

In the formula, (x is that (M, N are x to pixel, the pixel count of y direction for x, gray scale y) y) to Gray.

Described comentropy InEn is obtained by following formula:

$\mathrm{InEn}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}p\left(i\right){\mathrm{Log}}_{2}p\left(i\right)$

(7) according to average contrast C and comentropy InEn, obtain the value IQAF of evaluation function, described evaluation function is:

IQAF＝C×InEn

(8) tonal range Delta is added 1, turn back to step (five) again, terminal point up to the hunting zone;

(9) searched for after, find out the maximum IQAF and the Delta value of this IQAF correspondence, with the Zadeh-X transform method original-gray image is carried out conversion, obtain the one group of new T of grey scale pixel value as a result (x, y), (x, y) image of Gou Chenging is best quality image to this grey scale pixel value T.

Following table has shown the hunting zone according to delta, searches for the evaluation function value that obtains one by one:

As shown in Figure 3: among the figure, a) be original image shown in Figure 2, the image of correspondence when b) being Delta=8, the image of correspondence when c) being Delta=2 for the IQAF maximum.We as can be seen, near picture quality the best of IQAF maximum correspondence, and the average gray value G of this figure is 128.

Claims (5)

1, a kind of non-reference picture quality appraisement method based on comentropy and contrast is characterized in that carrying out as follows:

(1) obtains original image;

(2) original image is converted to original-gray image;

(3) (x, y) information are calculated the average gray value G of this original-gray image and gray scale spectrum, and obtain gray scale spectrum continuous bandwidth BW according to this gray scale spectrum to obtain the gray value 0 of this original-gray image;

(4) determine the hunting zone of gray scale initial value Theta and gray-level Delta;

(5) according to the gray value 0 of described gray scale initial value Theta, gray-level Delta and original image (x y), carries out conversion with the Zadeh-X transform method to original-gray image, obtain one group of new T of grey scale pixel value as a result (x, y);

(6) (x y), calculates this average contrast C and comentropy InEn of grey scale pixel value correspondence as a result according to the new T of grey scale pixel value as a result that obtains;

(7) according to average contrast C and comentropy InEn, obtain the value IQAF of evaluation function, described evaluation function is:

IQAF＝C×InEn

(8) gray-level Delta is added 1, turn back to step (five) again, up to the search terminal point;

(9), after search finishes, find out the maximum IQAF and the Delta value of this IQAF correspondence, with the Zadeh-X transform method original-gray image is carried out conversion, obtain one group of new T of grey scale pixel value as a result (x, y), this as a result grey scale pixel value T (x, y) image of Gou Chenging is best quality image.

2, according to the described non-reference picture quality appraisement method of claim 1 based on comentropy and contrast, it is characterized in that: in described step (four), described gray scale initial value Theta is the gray scale spectrum starting point gray value of band continuously, the starting point of gray-level Delta hunting zone is the average gray value G of original-gray image, and terminal point is

3, according to the described non-reference picture quality appraisement method based on comentropy and contrast of claim 2, it is characterized in that: the average gray value G of described gray level image is obtained by following formula:

$\stackrel{\‾}{G}=\frac{1}{M\×N}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\mathrm{Gray}(x,y)$

In the formula, (x is that (M, N are x to pixel, the pixel count of y direction for x, gray scale y) y) to Gray.

4, according to the described non-reference picture quality appraisement method based on comentropy and contrast of claim 1, it is characterized in that: in described step (six), described average contrast C is obtained by following formula:

$\stackrel{\‾}{C}=\frac{1}{(M-1)\×(N-1)}\underset{y=0}{\overset{N-2}{\mathrm{\Σ}}}\underset{x=0}{\overset{M-2}{\mathrm{\Σ}}}|\mathrm{Gray}(x,y)-\mathrm{Gray}(x+1,y)|$

In the formula, (x is that (M, N are x to pixel, the pixel count of y direction for x, gray scale y) y) to Gray.

5, according to the described non-reference picture quality appraisement method based on comentropy and contrast of claim 1, it is characterized in that: in described step (six), described comentropy worker nEn is obtained by following formula:

$\mathrm{InEn}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}p\left(i\right)L{\mathrm{og}}_{2}p\left(i\right)$

In the formula, gray scale is the probability of the pixel of i in p (i) the expression gray scale spectrum.

Images