CN105931206A - Method for enhancing sharpness of color image with color constancy - Google Patents

Abstract

The invention discloses a method for enhancing sharpness of color image with color constancy. The method includes the following steps: based on the red, green and blue three wave bands of an image, calculating a brightness image; setting a neighborhood radius, calculating mean value and standard difference of the brightness image in each pixel neighborhood, obtaining an neighborhood mean value image and a neighborhood standard difference image; based on the neighborhood mean value image and the neighborhood standard difference image, conducting adaptive calculation on weight of each pixel point; based on the neighborhood mean value image and the weights, recalculating a new brightness image; in accordance with the theory of color constancy, based on the new brightness image and an original brightness image, conducting same ratio enhancement on the three wave bands of the original image, obtaining the image with enhanced sharpness with color constancy. According to the invention, the method can greatly increase image sharpness, reduce color distortion as much as possible, maintain color constancy, and effectively increase visual effects of the image.

Description

A kind of coloured image definition enhancing method of color constancy

Technical field

The invention belongs to technical field of image processing, particularly relate to the coloured image definition of a kind of color constancy Enhancement Method.

Background technology

Owing to color-image forming apparatus is digital equipment, multimedia, biomedicine and being widely used in the Internet, Color image processing is increasingly subject to people's attention.Restrict due to the factor of each side or condition limit, The color image quality frequently resulted in may be undesirable, the most high situation of image blurring, definition occurs, Therefore before successive image processes, it is often necessary to coloured image is carried out enhancement process, to provide effective special Reference ceases, or improves its quality to obtain gratifying visual effect.

Coloured image is made up of gray level image, by the enhancing of gray level image, can reach coloured image The purpose strengthened.Gray level image strengthen main method such as: contrast stretching, amplitude limit, histogram equalization etc. In a lot of documents and materials, there is detail discussion, but these methods cannot be directly used to color image enhancement, Essentially consist in the chromatic value of pixel when of strengthening coloured image can not be changed, not so, the vision effect of image Fruit can change.The color image enhancement technology gray level image treatment technology that is far from is ripe.Coloured image Red, green, blue three primary colours also exist the highest dependency, must keep or recover, otherwise in enhancement process Chromatic distortion will be caused.

For keeping the problem of color enhancement image, there is a series of algorithm to propose, wherein involved little Wave conversion, Retinex are theoretical, and these algorithms preferably enhance image, and color keeps good, but Simultaneously as involve the computings such as the convolution of wavelet transformation and multiple Gaussian function, realizing good place While reason effect, pay bigger amount of calculation and longer operation time as cost, be poorly suitable for reality Real-time process in the work of border.

Color constancy refers to that the psychology not tended towards stability object color perception because of external environment change is inclined To.For coloured image, keep the color constancy of image, distortionless after being to ensure that image enhaucament Effective means.Early-stage Study finds, by strengthening three wave bands of coloured image on year-on-year basis, it is possible to meet The demand of color constancy, therefore, the present invention proposes the image definition enhancing method of a kind of color constancy.

Summary of the invention

Present invention seek to address that the problems such as definition is inadequate easily occurs in existing coloured image, and at definition The problem that cross-color easily occurs during enhancing, it is provided that the coloured image definition of a kind of color constancy increases Strong method.

Embodiments of the invention provide the coloured image definition enhancing method of a kind of color constancy, including: bag Include,

Three wave bands of RGB according to image calculate luminance picture；

Calculate luminance picture average on neighborhood and standard deviation, obtain neighboring mean value image and neighborhood standard deviation Image；

According to neighboring mean value image and neighborhood standard deviation image, calculate the weight of each pixel；

Calculate new luminance picture；

Calculate the result images that definition strengthens.

Three wave bands of RGB according to image calculate luminance picture method particularly includes:

I=(R+G+B)/3

Wherein, R, G, B are respectively three wave bands of RGB of image.

Further, concretely comprising the following steps of neighboring mean value image and neighborhood standard deviation image is calculated:

For each pixel on luminance picture, if the radius of neighbourhood is N, then Size of Neighborhood is (2N+1) * (2N+1), take out the value of its neighborhood territory pixel, calculate average and the standard deviation of this neighborhood, be assigned to this pixel, Respectively obtain the neighboring mean value image with original brightness image formed objects and neighborhood standard deviation image；

Calculate speed for improving, first luminance picture is carried out down-sampling M times, be then based on down-sampling luminance graph As calculating neighboring mean value image and neighborhood standard deviation image, then by neighboring mean value image and neighborhood standard deviation image Carrying out up-sampling 1/M times, wherein M≤1, up-sampling and Downsapling method can use nearest neighbor algorithm, bilinearity The methods such as interpolation, bicubic convolution method.

Further, according to neighboring mean value image and neighborhood standard deviation image, the weight of each pixel is calculated Method particularly includes:

$w(i,j)=\frac{mean\left(I\right)}{\sqrt{2*\mathrm{\π}*S(i,j)}*(S\_\max +S\_min)/2}*e^{-\frac{{\[\sqrt{S(i,j)}-(S\_max+S\_min)/2\]}^2}{2*{\[(S\_max+S\_min)/2\]}^2}}$

Wherein i and j represents pixel coordinate, and (i j) represents that (mean (I) represents calculating brightness to pixel for i, weights j) to w The average of image, (i j) represents that (i, j) value at place, S_max and S_min represents neighborhood standard deviation image in pixel to S The maximum of neighborhood standard deviation image and minima.

Further, the method for the luminance picture that described calculating is new is:

I_new (i, j)=Ave (i, j)+w (i, j) * [I (and i, j)-Ave (i, j)]

Wherein (i j) represents that (i, j) value at place, (i j) represents that luminance picture is in pixel to I to neighboring mean value image in pixel to Ave (i, j) value at place, (i j) represents pixel (i, j) weight at place to w.

Further, the result images that described calculating definition strengthens method particularly includes:

$\left\{\begin{array}{c}R\_new(i,j)=R(i,j)*I\_new(i,j)/I(i,j)\\ G\_new(i,j)=G(i,j)*I\_new(i,j)/I(i,j)\\ B\_new(i,j)=B(i,j)*I\_new(i,j)/I(i,j)\end{array}\right.$

Wherein R_new, G_new, B_new are three wave bands of the enhanced RGB of definition.

In above technical scheme, by calculating luminance picture neighboring mean value and neighborhood standard deviation, it is possible to catch figure As the texture strength feature of upper each pixel region, it is adaptively calculated the enhancing weight of each pixel accordingly, The self adaptation definition being capable of each pixel strengthens；During calculating neighboring mean value and neighborhood standard deviation, Image carries out first down-sampling up-sample again, it is possible to while ensureing to calculate effect, be effectively reduced computing Amount；According to color constancy principle, each pixel is strengthened on year-on-year basis such that it is able to be effectively prevented from color Color distortion, obtains good clarity enhancement effect.

Accompanying drawing explanation

Fig. 1 is the original image of mountain image；

Fig. 2 is that the histogram equalization method of mountain image strengthens image；

Fig. 3 is that the MSRCR method of mountain image strengthens image；

Fig. 4 is that the inventive method of mountain image strengthens image；

Fig. 5 is the original image of Flos Nelumbinis image；

Fig. 6 is that the histogram equalization method of Flos Nelumbinis image strengthens image；

Fig. 7 is that the MSRCR method of Flos Nelumbinis image strengthens image；

Fig. 8 is that the inventive method of Flos Nelumbinis image strengthens image；

Fig. 9 is the colour-image reinforcing method workflow schematic diagram of the color constancy of the embodiment of the present invention.

Detailed description of the invention

In order to make technical problem solved by the invention, technical scheme and beneficial effect clearer, with Lower combination drawings and Examples, are further elaborated to the present invention.Should be appreciated that described herein Specific embodiment only in order to explain the present invention, be not intended to limit the present invention.

The invention mainly comprises following step:

(1) luminance picture is calculated

The overall bright-dark degree of image has been reacted in the brightness of image, in multicolour space, has calculating bright The method of degree image, calculates by simplest mode here, takes the average of three wave bands of coloured image:

I=(R+G+B)/3

Wherein, R, G, B are respectively three wave bands of RGB of image.

(2) neighboring mean value image and neighborhood standard deviation image are calculated

For each pixel on luminance picture, if the radius of neighbourhood is N, then Size of Neighborhood is (2N+1) * (2N+1), take out the value of its neighborhood territory pixel, calculate average and the standard deviation of this neighborhood, be assigned to this pixel, Respectively obtain the neighboring mean value image with original brightness image formed objects and neighborhood standard deviation image；

Mean value computation formula is:

$Ave(i,j)=\frac{1}{{(2N+1)}^2}\underset{m=i-N}{\overset{i+N}{\mathrm{\Σ}}}\underset{n=j-N}{\overset{j+N}{\mathrm{\Σ}}}I(m,n)$

Standard deviation computing formula is:

$S(i,j)=\frac{1}{{(2N+1)}^2}\underset{m=i-N}{\overset{i+N}{\mathrm{\Σ}}}\underset{n=j-N}{\overset{j+N}{\mathrm{\Σ}}}{\[I(m,n)-Ave(m,n)\]}^2$

Wherein i and .j represents pixel coordinate, and N is the radius of neighbourhood, and Ave is neighboring mean value image, and S is neighborhood Standard deviation image.

Calculate speed for improving, first luminance picture is carried out down-sampling M times, be then based on down-sampling luminance graph As calculating neighboring mean value image and neighborhood standard deviation image, then by neighboring mean value image and neighborhood standard deviation image Carrying out up-sampling 1/M times, wherein M≤1, up-sampling and Downsapling method can use nearest neighbor algorithm, two-wire The property method such as interpolation, bicubic convolution method.

(3) weight of each pixel is calculated

The average of neighborhood and standard deviation have reacted the Local textural feature of image, it is possible to reflection image is clear indirectly Clear degree.On the basis of neighboring mean value image and neighborhood standard deviation image, calculate the weights of each pixel, For calculating the definition enhancing degree of each pixel:

$w(i,j)=\frac{mean\left(I\right)}{\sqrt{2*\mathrm{\π}*S(i,j)}*(S\_\max +S\_min)/r}*e^{-\frac{{\[\sqrt{S(i,j)}-(S\_max+S\_min)/2\]}^2}{2*{\[(S\_\max +S\_\min )/r\]}^2}}$

Wherein i and j represents pixel coordinate, and (i j) represents that (mean (I) represents calculating brightness to pixel for i, weights j) to w The average of image, (i j) represents that (i, j) value at place, S_max and S_min represents neighborhood standard deviation image in pixel to S The maximum of neighborhood standard deviation image and minima.

(4) new luminance picture is calculated

On the basis of definition weights, new luminance picture can be calculated by equation below:

I_new (i, j)=Ave (i, j)+w (i, j) * [I (and i, j)-Ave (i, j)]

Wherein (i j) represents that (i, j) value at place, (i j) represents that luminance picture is in pixel to I to neighboring mean value image in pixel to Ave (i, j) value at place, (i j) represents pixel (i, j) weight at place to w.

By above-mentioned calculating, definition higher brightness image can be obtained.

(5) result images that definition strengthens is calculated

In HSV space, the computational methods of H and S component are:

WhereinX₀=min (R, G, B), I=(R+G+B)/3, if before Chu Liing The color vector of certain pixel is X=(R, G, B), and ratio warp parameter is α, and translation parameters is β, then pass through Color vector after translation and ratio stretching is:

X '=(α R+ β, α G+ β, α B+ β)=α X+ β (2)

(1) formula that substituted into by X ' calculates and understands,Therefore H '=H.

Understand it addition, calculate S component,

$S^{\′}=\frac{I^{\′}-X_0^{\′}}{I^{\′}}=\frac{\mathrm{\α}(I-X_0)}{\mathrm{\α}I+\mathrm{\β}}---\left(3\right)$

If the translational movement β ≈ 0 on this pixel, then S ' ≈ S, the saturated of this pixel can be approximately considered Spend the most constant.In this case, tone and saturation are all not changed in, then the color letter of image after processing Breath will maximize the retention.It is to say, strengthened on year-on-year basis by pixel each to coloured image, I.e. can keep the constancy of color well.

Therefore, in the present invention, by new luminance picture and the ratio of original brightness image, calculate final Definition enhancing image:

$\left\{\begin{array}{c}R\_new(i,j)=R(i,j)*I\_new(i,j)/I(i,j)\\ G\_new(i,j)=G(i,j)*I\_new(i,j)/I(i,j)\\ B\_new(i,j)=B(i,j)*I\_new(i,j)/I(i,j)\end{array}\right.$

Wherein R_new, G_new, B_new are three wave bands of the enhanced RGB of definition.

In order to verify the effectiveness of Enhancement Method of the present invention, it is right that embodiments of the invention give many groups Than experiment, simultaneously by the method for the present invention and histogram equalization, multiple dimensioned Retinex color enhancement method (MSRCR) compare.Fig. 1 is the enhancing result of mountain image, and Fig. 2 is the enhancing knot of Flos Nelumbinis image Really.It can be seen that in FIG, the result of histogram equalization presents obvious color distortion, and MSRCR Result seem fuzzyyer.In fig. 2, the contrast of histogram equalization result is the most unbalance, and MSRCR There is the problem that color is the brightest.

For the effectiveness of objective evaluation each method, the present embodiment uses general image quality index to comment Valency.

Use articulation index evaluate strengthen image definition:

$D=\frac{1}{M\×N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\sqrt{\frac{{\[X(i,j)-X(i-1,j)\]}^2+{\[X(i,j)-X(i,j-1)\]}^2}{2}}$

For the contrast of result images and the brightness of each algorithm of objective evaluation, use conventional depending in existing research Average and the evaluation methodology of local mean variance according to image:

$C=\frac{1}{M\×N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\frac{|C_{Ori}(i,j)-C_{\mathrm{Re}}(i,j)|}{C_{Ori}(i,j)}$

$L=\frac{1}{M\×N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\frac{|L_{Ori}(i,j)-L_{\mathrm{Re}}(i,j)|}{L_{Ori}(i,j)}$

Wherein M and N is the length and width of image, C and L represents average and variance.In order to evaluate color Distortion level, according to These parameters, uses following tone and saturation deviation to evaluate:

$H=\frac{1}{M\×N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\frac{|H_{Ori}(i,j)-H_{\mathrm{Re}}(i,j)|}{H_{Ori}(i,j)}$

$S=\frac{1}{M\×N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\frac{|S_{Ori}(i,j)-S_{\mathrm{Re}}(i,j)|}{S_{Ori}(i,j)}$

Wherein H and S is respectively tone and the saturation of image.

According to evaluation index, D is the highest, and image definition is the highest；C, L, H, S are the least, and image color is also Former degree is the highest.

Evaluation result is as shown in the table:

The evaluation result of table 1. Fig. 1

The evaluation result of table 2. Fig. 2

	D	C	L	H	S
						Original image	1.9715
Histogram equalization	3.0622	0.4133	0.1513	0.1374	0.9239
						MSRCR	1.8842	0.3938	0.2594	0.1351	0.4712
The inventive method	3.9072	0.0161	8.2939e-005	0.0017	0.0077

Although as it can be seen from table 1 the definition of histogram equalizing method is the highest, but C, L, H, S Desired value all much larger than the method for the present invention, it can be seen that histogram equalizing method from result images Fine definition is with color distortion as cost, and the inventive method is in addition to definition is less than histogram equalization method, Remaining index is the most excellent with histogram equalization method and MSRCR.Similarly, in table 2, the inventive method exists The most excellent and histogram equalization method and MSRCR in the index such as definition, C, L, H, S.Consolidated statement 1 He The result of table 2 is it can be seen that the inventive method has the highest definition, and color distortion degree is the most remote simultaneously Less than histogram equalization and MSRCR method, show that the inventive method is improving while image definition, Color distortion can be effectively prevented from.

The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all at this Any amendment, equivalent and the improvement etc. made within bright spirit and principle, should be included in the present invention Protection domain within.

Claims (6)

1. the coloured image definition enhancing method of a color constancy, it is characterised in that: include,

Three wave bands of RGB according to image calculate luminance picture；

Calculate luminance picture average on neighborhood and standard deviation, obtain neighboring mean value image and neighborhood standard deviation Image；

According to neighboring mean value image and neighborhood standard deviation image, calculate the weight of each pixel；

Calculate new luminance picture；

Calculate the result images that definition strengthens.

The coloured image definition enhancing method of color constancy the most according to claim 1, its feature exists In: described three wave bands of RGB according to image calculate luminance picture method particularly includes:

I=(R+G+B)/3

Wherein, R, G, B are respectively three wave bands of RGB of image.

The coloured image definition enhancing method of color constancy the most according to claim 1, its feature exists In: calculate concretely comprising the following steps of neighboring mean value image and neighborhood standard deviation image:

For each pixel on luminance picture, if the radius of neighbourhood is N, then Size of Neighborhood is (2N+1) * (2N+1), take out the value of its neighborhood territory pixel, calculate average and the standard deviation of this neighborhood, be assigned to this pixel, Respectively obtain the neighboring mean value image with original brightness image formed objects and neighborhood standard deviation image；

Calculate speed for improving, first luminance picture is carried out down-sampling M times, be then based on down-sampling luminance graph As calculating neighboring mean value image and neighborhood standard deviation image, then by neighboring mean value image and neighborhood standard deviation image Carrying out up-sampling 1/M times, wherein M≤1, up-sampling and Downsapling method can use nearest neighbor algorithm, bilinearity The methods such as interpolation, bicubic convolution method.

The coloured image definition enhancing method of color constancy the most according to claim 1, its feature exists In: according to neighboring mean value image and neighborhood standard deviation image, calculate the concrete grammar of the weight of each pixel For:

$w(i,j)=\frac{mean\left(I\right)}{\sqrt{2*\mathrm{\π}*S(i,j)}*(S\_\max +S\_min)/2}*e^{-\frac{{\[\sqrt{S(i,j)}-(S\_max+S\_min)/2\]}^2}{2*{\[(S\_max+S\_min)/2\]}^2}}$

Wherein i and j represents pixel coordinate, and (i j) represents that (mean (I) represents calculating brightness to pixel for i, weights j) to w The average of image, (i j) represents that (i, j) value at place, S_max and S_min represents neighborhood standard deviation image in pixel to S The maximum of neighborhood standard deviation image and minima.

The coloured image definition enhancing method of color constancy the most according to claim 1, its feature exists In: the method for the luminance picture that described calculating is new is:

I_new (i, j)=Ave (i, j)+w (i, j) * [I (and i, j)-Ave (i, j)]

Wherein (i j) represents that (i, j) value at place, (i j) represents that luminance picture is in pixel to I to neighboring mean value image in pixel to Ave (i, j) value at place, (i j) represents pixel (i, j) weight at place to w.

The coloured image definition enhancing method of color constancy the most according to claim 1, its feature exists In: the result images that described calculating definition strengthens method particularly includes:

$\left\{\begin{array}{c}R\_new(i,j)=R(i,j)*I\_new(i,j)/I(i,j)\\ G\_new(i,j)=G(i,j)*I\_new(i,j)/I(i,j)\\ B\_new(i,j)=B(i,j)*I\_new(i,j)/I(i,j)\end{array}\right.$

Wherein R_new, G_new, B_new are three wave bands of the enhanced RGB of definition.