CN101625759A - Image quality evaluation method - Google Patents

Abstract

The invention discloses an image quality evaluation method which is characterized by comprising the following steps: (1) selecting a gray level image as an evaluated image; (2) acquiring the gray value of each pixel of the evaluated image; (3) calculating the comentropy InEn, the average contrast AC, the normalization gray difference NGD and the key region gray standard difference SDKR of the evaluated image; (4) establishing the gray image quality evaluation function NCAF; and (5) calculating the value of the gray image quality evaluation function NCAF of the evaluated image. The invention can evaluate the evaluated image without depending on a reference image, the evaluation result satisfies the human vision subjective recognition, and the larger the NCAF value is, the better the quality of the evaluated image is.

Description

Image quality evaluating method

Technical field

The present invention relates to image processing field, specifically, is a kind of gray scale method for quality that is used to estimate.

Background technology

Image quality evaluation is to estimate the basis of the identification of imaging device quality, imaging device monitoring and image and classification and the key of decision-making, has important military, safety and civilian meaning.A variety of methods are arranged aspect Flame Image Process at present,, need relatively to handle the picture quality that obtains, select the quality better image through different images as histogram equalization, contrast stretching etc.

Image quality evaluation mainly is divided into: full reference image quality appraisement, non-reference picture quality appraisement, simplify reference image quality appraisement.And the progress of image quality evaluation at present is mainly at the full reference image quality appraisement of gray level image, and the main evaluation map picture situation that degrades after treatment, all need to look like to carry out calculated crosswise, do not utilize the quality assessment of large-scale image to reference picture with by evaluation map.

Summary of the invention

The object of the present invention is to provide a kind of image quality evaluating method, can under the situation that does not need reference picture, estimate the quality of grayscale image quality.

To achieve these goals, technical scheme of the present invention is as follows: a kind of image quality evaluating method, its key are to carry out as follows:

(1) selecting gray level image is by the evaluation map picture;

(2) obtain by each gray values of pixel points of evaluation map picture;

(3) calculate by the information entropy InEn of evaluation map picture, average contrast AC, Normalized Grey Level difference NGD and critical area gray standard deviation SD _KR

Described information entropy 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 that p (i) expression is distributed as pixel count on the i gray level by evaluation map when p (i)=0, makes Log ₂P (i)=0;

Described average contrast AC is obtained by following formula:

$\mathrm{AC}=\frac{1}{\sqrt{2}}\sqrt{{\mathrm{AC}}_x^2+{\mathrm{AC}}_y^2}$

Wherein, AC _x, AC _yRepresent respectively by the average contrast of evaluation map as X, Y direction, AC _x, AC _yComputing formula be respectively:

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

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

In last two formulas, (x is that (M, N are by the pixel count of evaluation map picture in X, Y direction for x, gray scale y) as pixel by evaluation map y) to Gray;

Described Normalized Grey Level difference NGD is obtained by following formula:

$\mathrm{NGD}=\frac{\mathrm{AOG}-|\mathrm{AOG}-\mathrm{AG}|}{\mathrm{AOG}}$

In the following formula, || be the absolute value operator, AOG represents the best average gray of human vision, calculates to such an extent that its value is 127.5 according to the even distribution histogram of ideal; AG represents by evaluation map to be calculated by following formula as average gray value:

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

In the following formula, (x is that (x, gray-scale value y), M, N are by the pixel count of evaluation map picture in X, Y direction as pixel by evaluation map y) to Gray;

Described is by background homogeneous area in the evaluation map picture by the critical area of evaluation map picture, critical area standard deviation SD _KRObtain by following formula:

${\mathrm{SD}}_{\mathrm{KR}}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-\mathrm{\μ})}^2}$

In the following formula, x _iExpression critical area picture is counted the gray-scale value of point, and μ represents the average of critical area pixel gray-scale value;

(4) set up grayscale image quality evaluation function NCAF, used formula is:

$\mathrm{NCAF}=\frac{{\mathrm{InEn}}^{\mathrm{\α}}{\mathrm{AC}}^{\mathrm{\β}}{\mathrm{NGD}}^{\mathrm{\γ}}}{1+{\mathrm{SD}}_{\mathrm{KR}}^{\mathrm{\η}}}$

Consideration is during by the noise of evaluation map picture, α=1, β=1, γ=1, η ∈ [0.8,1.0];

When not considering by the noise of evaluation map picture, α=1, β=1, γ=1, ${\mathrm{SD}}_{\mathrm{KR}}^{\mathrm{\η}}=0;$

(5) calculate by the value of the grayscale image quality evaluation function NCAF of evaluation map picture, the NCAF value is big more, and is good more by the quality of evaluation map picture.

This evaluation function is characterized as the basis with human vision to the cognitive function of image, also can be described as the mathematical model of human vision image quality evaluation.At first, do not have suitable illuminance, just can not get the good image of quality, it is exactly suitable gray scale that suitable scene illumination is reflected in the image. Secondly, the measured image of matter has suitable gray scale or chrominance information, and is promptly relevant with information entropy; The 3rd, the measured image of matter has the suitable space distribution of suitable gray scale or chrominance information, and is promptly relevant with contrast; The 4th, the measured image of matter has the low noise level of trying one's best, and is promptly relevant with standard deviation.Wherein colourity is for coloured image.

But gray scale, information entropy, contrast, noise can be described as the physical parameter of four basic objective measurements of picture quality description, i.e. picture quality four parameters.Can see that first three parameter all is suitable, the 4th parametric noise, it is low to try one's best.Find in the practice, when grayscale image quality evaluation function NCAF value is maximum, image best in quality.

Picture noise is gradation of image or near the fluctuation of chromatic value average.In electrocardiosignal, we observe noise at the equipotential line place, rather than remove to observe noise changing violent R ripple place, therefore, to correctly reflect picture noise, should select the background homogeneous area in the image to measure, promptly estimate the noise of entire image with image critical area noise, that is to say that critical area gradation of image standard deviation is to be used for a parameter of measurement image noise.The image critical area define dual mode: first kind, a manually selected background uniformly the zone as the image critical area; Second kind, the size of elder generation's definition image critical area, the standard deviation of in by the evaluation map picture, searching for the zone identical then with image critical area size, the zone of standard deviation minimum is the image critical area.

When considering picture noise, α=1, β=1, γ=1, η ∈ [0.8,1.0], the grayscale image quality evaluation function becomes:

$\mathrm{NCAF}=\frac{\mathrm{InEn}\×\mathrm{AC}\×\mathrm{NGD}}{1+{\mathrm{SD}}_{\mathrm{KR}}^{\mathrm{\η}}}$

When not considering picture noise, α=1, β=1, γ=1, ${\mathrm{SD}}_{\mathrm{KR}}^{\mathrm{\η}}=0,$ Grayscale image quality evaluation function NCAF becomes:

NCAF＝InEn×AC×NGD

Beneficial effect: can be according to the cognitive function feature of human vision to image, utilize picture quality four parameters: gray scale, information entropy, contrast, noise are set up grayscale image quality evaluation function NCAF, estimate the quality of the gray level image of any spectrum distribution.Compare with existing full reference image quality appraisement, this evaluation function has following advantage: do not rely on the mutual calculating of reference picture; Help the quality assessment of large-scale image; Can realize the comparison of different sized images quality; But the quality of evaluation reference image itself does not need reference picture is made the supposition of priori.

Description of drawings

Fig. 1 is a schematic flow sheet of the present invention;

Fig. 2 is the source images of embodiment;

Fig. 3 is the image of Fig. 2 through obtaining after the different grey scale change;

Wherein, Fig. 3 (a) is the image that obtains after Fig. 2 stretches through contrast; Fig. 3 (b) is the image that obtains after Fig. 2 moves to right through gray scale; Fig. 3 (c) is that Fig. 2 adds the image that obtains after the Gaussian noise; Fig. 3 (d) is the image of Fig. 2 through obtaining after the Fuzzy Processing.

Embodiment

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

Embodiment:

As shown in Figure 1: a kind of image quality evaluating method is characterized in that carrying out as follows:

(1) selecting gray level image is by the evaluation map picture;

(2) obtain by each gray values of pixel points of evaluation map picture;

(3) calculate by the information entropy InEn of evaluation map picture, average contrast AC, Normalized Grey Level difference NGD and critical area gray standard deviation SD _KR

Described information entropy 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 that p (i) expression is distributed as pixel count on the i gray level by evaluation map when p (i)=0, makes Log ₂P (i)=0;

Described average contrast AC is obtained by following formula:

$\mathrm{AC}=\frac{1}{\sqrt{2}}\sqrt{{\mathrm{AC}}_x^2+{\mathrm{AC}}_y^2}$

Wherein, AC _x, AC _yRepresent respectively by the average contrast of evaluation map as X, Y direction, AC _x, AC _yComputing formula be respectively:

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

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

In last two formulas, (x is that (M, N are by the pixel count of evaluation map picture in X, Y direction for x, gray scale y) as pixel by evaluation map y) to Gray;

Described Normalized Grey Level difference NGD is obtained by following formula:

$\mathrm{NGD}=\frac{\mathrm{AOG}-|\mathrm{AOG}-\mathrm{AG}|}{\mathrm{AOG}}$

In the following formula, || be the absolute value operator, AOG represents the best average gray of human vision, calculates to such an extent that its value is 127.5 according to the even distribution histogram of ideal; AG represents by evaluation map to be calculated by following formula as average gray value:

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

In the following formula, (x is that (x, gray-scale value y), M, N are by the pixel count of evaluation map picture in X, Y direction as pixel by evaluation map y) to Gray;

Described is by background homogeneous area in the evaluation map picture by the critical area of evaluation map picture, critical area standard deviation SD _KRObtain by following formula:

${\mathrm{SD}}_{\mathrm{KR}}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-\mathrm{\μ})}^2}$

In the following formula, x _iExpression critical area picture is counted the gray-scale value of point, and μ represents the average of critical area pixel gray-scale value;

(4) set up grayscale image quality evaluation function NCAF, used formula is:

$\mathrm{NCAF}=\frac{\mathrm{InEn}\×\mathrm{AC}\×\mathrm{NGD}}{1+{\mathrm{SD}}_{\mathrm{KR}}^{\mathrm{\η}}}$

Consideration is during by the noise of evaluation map picture, α=1, β=1, γ=1, η ∈ [0.8,1.0];

When not considering by the noise of evaluation map picture, α=1, β=1, γ=1, ${\mathrm{SD}}_{\mathrm{KR}}^{\mathrm{\η}}=0;$

(5) calculate by the value of the grayscale image quality evaluation function NCAF of evaluation map picture, the NCAF value is big more, and is good more by the quality of evaluation map picture.

The present invention utilizes document " modern image quality evaluation " [Zhou Wang and Alan C.Bovik.ModernImage Quality Assessment.Morgan ﹠amp; Claypool Publishers.San Rafael, CA USA.2006.] in einstein's head portrait of providing as source images, as shown in Figure 2, source images shown in Figure 2 carried out different grey scale change after, obtain one group of gray level image after changing, as shown in Figure 3.Fig. 2 and image shown in Figure 3 as by the evaluation map picture, are carried out quality assessment to them.

When considering by the noise of evaluation map picture, α=1, β=1, γ=1, η ∈ [0.8,1.0] gets η=0.8,0.9, respectively at 1.0 o'clock, calculates by the quality assessment function NCAF value of evaluation map picture, and the result is as shown in table 1.

In the present embodiment, by evaluation map picture the right collar place, artificial selection upper left corner coordinate be 10 * 16 the blockage conduct of (160,203) by the critical area of evaluation map picture, the gray standard deviation of the critical area of Fig. 2 and image shown in Figure 3 sees Table 1 the 5th row.The InEn value of Fig. 2 and image shown in Figure 3, AC value, AG value, NCAF value are shown in Table 1.As can be seen, Fig. 3 (a) has higher contrast than Fig. 2, and Fig. 3 (c) compares Fig. 2, Fig. 3 (a), Fig. 3 (b), Fig. 3 (d), and more noise is arranged; Fig. 2, Fig. 3 (a), Fig. 3 (b), Fig. 3 (c) compare Fig. 3 (d), and visual effect is better.

Five width of cloth are seen Table 1 last row by the ordering of the NCAF value of evaluation map picture, can see, the NCAF value of Fig. 3 (a) is maximum, and ordering the first shows the best in quality of it, and the NCAF value of Fig. 3 (d) is minimum, sorts the 5th, shows that its quality is the poorest.This also meets the subjective understanding result of human vision.

Table 1

By the evaluation map picture	??InEn	??AC	??AG	??SD KR	??NCAF ??(η＝0.8)	??NCAF ??(η＝0.9)	??NCAF ??(η＝1.0)	The NCAF ordering
									Fig. 2	??7.2585	??6.2641	??116.2500	??3.7976	??10.6077	??9.5892	??8.6410	??3
Fig. 3 (a)	??7.6563	??8.8287	??122.4045	??3.5435	??17.3447	??15.7416	??14.2827	??1
									Fig. 3 (b)	??7.2501	??6.1943	??134.3198	??3.5472	??11.3243	??10.3040	??9.3480	??2
Fig. 3 (c)	??7.4536	??14.0477	??116.6749	??15.7127	??9.5216	??7.4082	??5.7311	??4
									Fig. 3 (d)	??6.9749	??2.2773	??116.4450	??7.3875	??2.4372	??2.0582	??1.7296	??5

When not considering by the noise of evaluation map picture, α=1, β=1, γ=1, ${\mathrm{SD}}_{\mathrm{KR}}^{\mathrm{\η}}=0,$ Grayscale image quality evaluation function NCAF becomes:

NCAF＝InEn×AC×NGD

To being seen Table 2 by the InEn of evaluation map picture, AC, AG, NCAF and NCAF ordering shown in Fig. 2 and Fig. 3 (a), Fig. 3 (b), Fig. 3 (d).Because Fig. 3 (c) is the image after Fig. 2 adds Gaussian noise, therefore when not considering by the noise of evaluation map picture, Fig. 3 (c) is the same with the evaluation result of Fig. 2, is not therefore considering by the noise of evaluation map picture, Fig. 3 (c) is not done quality assessment.

Table 2

By the evaluation map picture	??InEn	??AC	??AG	??NCAF	The NCAF ordering
						Fig. 2	??7.2585	??6.2641	??116.2500	??41.4561	??3
Fig. 3 (a)	??7.6563	??8.8287	??122.4045	??64.8933	??1
						Fig. 3 (b)	??7.2501	??6.1943	??134.3198	??42.5070	??2
Fig. 3 (d)	??6.9749	??2.2773	??116.4450	??14.5069	??4

As can be seen from Table 2, when not considering by the noise of evaluation map picture, the NCAF value of Fig. 3 (a) is maximum, and is best in quality, and the NCAF value of Fig. 3 (d) is minimum, and quality is the poorest.This also meets the visual recognition of people to Fig. 2, Fig. 3 (a), Fig. 3 (b), Fig. 3 (d).

Claims (1)

1, a kind of image quality evaluating method is characterized in that carrying out as follows:

(1) selecting gray level image is by the evaluation map picture;

(2) obtain by each gray values of pixel points of evaluation map picture;

(3) calculate by the information entropy InEn of evaluation map picture, average contrast AC, Normalized Grey Level difference NGD and critical area gray standard deviation SD _KR

Described information entropy InEn is obtained by following formula:

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

In the formula, the probability that p (i) expression is distributed as pixel count on the i gray level by evaluation map when p (i)=0, makes Log ₂P (i)=0;

Described average contrast AC is obtained by following formula:

$\mathrm{AC}=\frac{1}{\sqrt{2}}\sqrt{{\mathrm{AC}}_x^2+{\mathrm{AC}}_y^2}$

Wherein, AC _x, AC _yRepresent respectively by the average contrast of evaluation map picture in X, Y direction, AC _x, AC _yComputing formula be respectively:

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

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

In last two formulas, (x is that (M, N are by the pixel count of evaluation map picture in X, Y direction for x, gray scale y) as pixel by evaluation map y) to Gray;

Described Normalized Grey Level difference NGD is obtained by following formula:

$\mathrm{NGD}=\frac{\mathrm{AOG}-|\mathrm{AOG}-\mathrm{AG}|}{\mathrm{AOG}}$

In the following formula, || be the absolute value operator, AOG represents the best average gray of human vision; AG represents by evaluation map to be calculated by following formula as average gray value:

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

In the following formula, (x is that (x, gray-scale value y), M, N are by the pixel count of evaluation map picture in X, Y direction as pixel by evaluation map y) to Gray;

Described is by background homogeneous area in the evaluation map picture by the critical area of evaluation map picture, critical area standard deviation SD _KRObtain by following formula:

${\mathrm{SD}}_{\mathrm{KR}}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-\mathrm{\μ})}^2}$

In the following formula, x _iExpression critical area picture is counted the gray-scale value of point, and μ represents the average of critical area pixel gray-scale value;

(4) set up grayscale image quality evaluation function NCAF, used formula is:

$\mathrm{NCAF}=\frac{\mathrm{InE}{n}^{\mathrm{\α}}{\mathrm{AC}}^{\mathrm{\β}}{\mathrm{NGD}}^{\mathrm{\γ}}}{1+{\mathrm{SD}}_{\mathrm{KR}}^{\mathrm{\η}}}$

Consideration is during by the noise of evaluation map picture, α=1, β=1, γ=1, η ∈ [0.8,1.0];

When not considering by the noise of evaluation map picture, α=1, β=1, γ=1, ${\mathrm{SD}}_{\mathrm{KR}}^{\mathrm{\η}}=0;$

(5) calculate by the value of the grayscale image quality evaluation function NCAF of evaluation map picture.

Images