CN104616259A - Non-local mean image de-noising method with noise intensity self-adaptation function - Google Patents

Abstract

The invention discloses a non-local mean image de-noising method with a noise intensity self-adaptation function. The non-local mean image de-noising method with the noise intensity self-adaptation function comprises the steps that firstly, a gray scale strip image is collected, and de-noising processing is conducted on the gray scale strip image by using different de-noising intensity parameters through a non-local mean method, so that optimal de-noising intensity parameters under the condition of different degrees of brightness are obtained; then, optimal de-noising intensity parameters corresponding to other degrees of brightness are calculated through a linear interpolation method; finally, de-noising processing is conducted on the image in a logarithm domain through the optimal de-noising intensity parameters corresponding to different degrees of brightness, and exponential transformation is conducted on the de-noised image in the logarithm domain, so that a final de-noised image is obtained. The non-local mean image de-noising method overcomes the defect that in an existing method, de-noising intensity parameters are fixed, and the de-noising effect of the image is improved; due to processing in the logarithm domain, the difference of brightness of pixels in a dark region can be increased, the difference of brightness of pixels in a bright region can be reduced, and the de-noising effect of the image can be improved.

Description

The adaptive non-local mean image de-noising method of a kind of noise intensity

Technical field

The invention belongs to digital image processing techniques field, relate to the adaptive non-local mean image de-noising method of a kind of noise intensity.

Background technology

Digital picture is in the process obtained, inevitably be subject to the interference of various noise signal, make deteriroation of image quality, thus affect the image characteristics extraction in later stage, Target Segmentation and target identification, therefore image denoising has important actual application value.

Image de-noising method can be divided into the method based on spatial domain and the large class of the method based on transform domain two.Method based on spatial domain has the method such as bilateral filtering, gaussian filtering based on single pixel grey scale similarity, based on the method for transform domain as various based on the image de-noising method etc. of wavelet transformation.Traditional spatial domain denoising method processes based on single Pixel Information, weak edge and grain details can not be retained well, the non-local mean denoising method proposed by Buades is then the information utilizing topography's block, better can express the structural information of image, therefore its performance is better than the denoise algorithm of other classics, as bilateral filtering, PDE, method etc. based on small echo.

Because non-local mean method has that algorithm is succinct, superior performance, is easy to improve and extend, it is a kind of main stream approach in current practical application.But the method is when practical application, identical removing-noise strength parameter is adopted to entire image, cause the denoising effect of different luminance area in image not ideal enough.The present invention is intended to distribute inconsistent feature according to the different luminance area noise intensity of image, by the denoising effect under different brightness in test grayscale bar, select the denoising parameter of varying strength, better can adapt to the uneven situation of noise profile, thus can obtain better image denoising effect.

Summary of the invention

The object of this invention is to provide the adaptive non-local mean image de-noising method of a kind of noise intensity, to solve the undesirable technical matters of image denoising effect that existing non-local mean denoising method adopts identical removing-noise strength parameter to cause.

The technical solution used in the present invention is, the adaptive non-local mean image de-noising method of a kind of noise intensity, specifically comprises following methods step:

Step 1: gather GTG bar image, inputted computing machine, GTG bar image is designated as z (i), and wherein i represents pixel, and z represents the brightness value of this pixel, and luminance areas different for GTG bar is designated as X _m;

Step 2: obtain the best removing-noise strength parameter under different brightness;

2.1, use non-local mean method under different removing-noise strength parameter, to carry out denoising to GTG bar image z (i), to obtain different brightness Y _mcorresponding best removing-noise strength parameter g _m, be designated as (g _m, Y _m), Y _mfor luminance area X _maverage brightness;

2.2, adopt linear difference method to obtain and be different from brightness Y _mbrightness P _ncorresponding best removing-noise strength parameter q _n;

Step 3: best denoising parameter corresponding to the different brightness obtained according to step 2 treats the noise intensity self-adaptive solution that denoising image carries out under different brightness.

Feature of the present invention is also,

The detailed process of step 2.1 is:

2.1.1, adopt denoising formula to GTG bar image z (i) denoising:

By removing-noise strength parameter h from 1, change value successively from small to large, carry out denoising to GTG bar image z (i), obtain the image after a series of denoising, wherein, the value of removing-noise strength parameter h is: h ₁=1, h _i=10 (i-1), 2≤i≤101, i is integer, h _irepresent i-th removing-noise strength parameter, denoising formula is as follows:

NLM(i)＝∑ω(i,j)z(j)

$\mathrm{\ω}(i,j)=\frac{1}{C\left(i\right)}\exp (-\frac{{\left|\right|z\left(N_i\right)-z\left(N_j\right)\left|\right|}_2^2}{h^2})$

$C\left(i\right)=\underset{j}{\mathrm{\Σ}}\exp (-\frac{{\left|\right|z(N_i-z\left(N_j\right))\left|\right|}_2^2}{h^2});$

Wherein, NLM (i) is the image after gray scale image z (i) being used the denoising of non-local mean method, and ω (i, j) represents the weight between pixel i and j, and meet 0≤ω (i, j)≤1 and j is the pixel in 21 × 21 regions centered by i; C (i) is normalized factor, N _irepresent the image block of 7 × 7 centered by pixel i, N _jrepresent the image block of 7 × 7 centered by pixel j;

Step 2.1.2, calculates GTG bar image z (i) each luminance area X _madopt different removing-noise strength parameter h _ithe Y-PSNR PSNR of a series of images obtained after denoising, chooses each luminance area X _mremoving-noise strength parameter h corresponding to the image that in a series of images obtained after denoising, Y-PSNR PSNR is maximum _ias corresponding bright region X _mbest removing-noise strength parameter, be designated as g _m, and calculate each luminance area X _maverage brightness Y _m, by each luminance area X _munder best removing-noise strength parameter and average brightness be expressed as (g _m, Y _m);

The computing formula of Y-PSNR PSNR is:

$\mathrm{PSNR}=10\log \left(\frac{{255}^2}{\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{(\mathrm{NLM}\left(i\right)-z\left(i\right))}^2}\right);$

Wherein, M represents luminance area X _msum of all pixels;

The average brightness Y of each luminance area _mcomputing formula is:

$Y_m=\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}z\left(i\right).$

The detailed process of step 2.2 is:

If brightness P _nnot at (g _m, Y _m) among, then from (g _m, Y _m) among find adjacent two average brightness the most close with its numerical value, be designated as Y respectively _jand Y _j+1, wherein Y _jbe less than P _n, Y _j+1be greater than P _n, Y _jand Y _j+1at (g _m, Y _m) in corresponding removing-noise strength parameter be respectively g _jand g _j+1, then brightness P _ncorresponding best removing-noise strength parameter q _ntried to achieve by linear interpolation formulae discovery, linear interpolation formula is:

$q_n=\frac{|P_n-Y_j|}{|Y_{j+1}-Y_j|}\×g_j+\frac{|Y_{j+1}-P_n|}{|Y_{j+1}-Y_j|}\×g_{j+1}.$

The detailed process of step 3 is:

3.1, by treating that the image of denoising is expressed as g (x, y), carry out natural logarithm conversion to it, transformation results is g ₁(x, y)=lng (x, y);

3.2, at log-domain to g ₁(x, y) adopts non-local mean method to carry out denoising, is specially:

To g ₁each pixel (x, y) of (x, y), adopts the removing-noise strength parameter h of the best that its original image g (x, y) brightness value is corresponding, according to denoising formula to g ₁(x, y) denoising, obtains the image after denoising, is designated as g ₂(x, y);

3.3, to g ₂(x, y) carries out exponential transform, obtains the image h (x, y) after final denoising, and result is h (x, y)=exp (g ₂(x, y).

The invention has the beneficial effects as follows, the present invention is by adopting different removing-noise strength parameters to carry out non-local mean denoising to GTG bar image, seek the best removing-noise strength parameter under different brightness, then different removing-noise strength parameter denoisings is adopted to image, overcome the defect that in existing non-local mean denoising method, removing-noise strength parameter is fixing, improve the denoising effect of the different luminance area of image.Meanwhile, contribute in log-domain process the difference increasing dark areas pixel intensity, reduce the difference of bright area, improve the denoising effect of image further.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the adaptive non-local mean image de-noising method of a kind of noise intensity of the present invention;

Fig. 2 is the image treating denoising;

Fig. 3 is the image adopting the inventive method denoising later.

Embodiment

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

The invention provides the adaptive non-local mean image de-noising method of a kind of noise intensity, specifically implement according to following steps:

Step 1: gather the GTG bar image on KODAK Gray Scale GTG card, inputted computing machine, this GTG bar image comprises the region from black to white totally 20 different brightness, GTG bar image is designated as z (i), wherein i represents pixel, and z represents the brightness value of this pixel, and luminance area is designated as X _m, 1≤m≤20;

Step 2: obtain the best removing-noise strength parameter under different brightness, detailed process is as follows:

Step 2.1, uses non-local mean algorithm, uses different removing-noise strength parameters to carry out denoising, to obtain different brightness Y to gray scale image z (i) _mcorresponding best removing-noise strength parameter g _m, be designated as (g _m, Y _m| m=1,2......20), Y _mfor luminance area X _maverage brightness:

Step 2.1.1, adopts denoising formula to GTG bar image z (i) denoising:

By removing-noise strength parameter h from 1, change value successively from small to large, carry out denoising to GTG bar image z (i), obtain the image after a series of denoising, wherein, the value of removing-noise strength parameter h is: h ₁=1, h _i=10 (i-1), 2≤i≤101, i is integer, h _irepresent the value of i-th removing-noise strength parameter, denoising formula is as follows:

NLM(i)＝∑ω(i,j)z(j) (1)

$\mathrm{\ω}(i,j)=\frac{1}{C\left(i\right)}\exp (-\frac{{\left|\right|z\left(N_i\right)-z\left(N_j\right)\left|\right|}_2^2}{h^2})---\left(2\right)$

$C\left(i\right)=\underset{j}{\mathrm{\Σ}}\exp (-\frac{{\left|\right|z(N_i-z\left(N_j\right))\left|\right|}_2^2}{h^2});---\left(3\right)$

Wherein, NLM (i) is the image after gray scale image z (i) being used the denoising of non-local mean method, and ω (i, j) represents the weight between pixel i and j, and meet 0≤ω (i, j)≤1 and j is the pixel in 21 × 21 regions centered by i; C (i) is normalized factor, N _irepresent the image block of 7 × 7 centered by pixel i, N _jrepresent the image block of 7 × 7 centered by pixel j;

Step 2.1.2, calculates GTG bar image z (i) each luminance area X _madopt different removing-noise strength parameter h _ithe Y-PSNR PSNR of a series of images obtained after denoising, chooses each luminance area X _mremoving-noise strength parameter h corresponding to the image that in a series of images obtained after denoising, Y-PSNR PSNR is maximum _ias corresponding bright region X _mbest removing-noise strength parameter, be expressed as g _m, and calculate each luminance area X _maverage brightness Y _m, by each luminance area X _munder best removing-noise strength parameter and average brightness be expressed as (g _m, Y _m| m=1,2......20);

The computing formula of Y-PSNR PSNR is:

$\mathrm{PSNR}=10\log \left(\frac{{255}^2}{\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{(\mathrm{NLM}\left(i\right)-z\left(i\right))}^2}\right)---\left(4\right)$

Wherein, M represents luminance area X _msum of all pixels;

The average brightness computing formula of each luminance area is:

$Y_m=\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}z\left(i\right)---\left(5\right)$

Wherein, M represents luminance area X _msum of all pixels;

Step 2.2, adopts linear difference method to obtain and is different from brightness Y _mbrightness P _ncorresponding best removing-noise strength parameter q _n, be specially:

If brightness P _nnot at (g _m, Y _m| m=1,2......20) among, then from (g _m, Y _m| m=1,2......20) among find adjacent two average brightness the most close with its numerical value, be designated as Y respectively _jand Y _j+1, wherein Y _jbe less than P _n, Y _j+1be greater than P _n, Y _jand Y _j+1at (g _m, Y _m| m=1,2......20) in corresponding removing-noise strength parameter be respectively g _jand g _j+1, then brightness P _ncorresponding best removing-noise strength parameter q _ntried to achieve by linear interpolation formulae discovery, linear interpolation formula is:

$q_n=\frac{|P_n-Y_j|}{|Y_{j+1}-Y_j|}\×g_j+\frac{|Y_{j+1}-P_n|}{|Y_{j+1}-Y_j|}\×g_{j+1}---\left(6\right)$

Step 3, best denoising parameter corresponding to the different brightness obtained according to step 2 treats the noise intensity self-adaptive solution that denoising image carries out under different brightness, and detailed process is:

Step 3.1, by treating that the image of denoising is expressed as g (x, y), carries out natural logarithm conversion to it, and transformation results is g ₁(x, y)=lng (x, y);

Step 3.2, at log-domain to g ₁(x, y) adopts non-local mean method to carry out denoising, is specially:

To g ₁each pixel (x, y) of (x, y), adopts the removing-noise strength parameter h of the best that its original image g (x, y) brightness value is corresponding, according to the denoising formula in step 2 to g ₁(x, y) denoising, obtains the image after denoising, is designated as g ₂(x, y);

Step 3.3), to g ₂(x, y) carries out exponential transform, obtains the image h (x, y) after final denoising, and result is h (x, y)=exp (g ₂(x, y)).

Adopt GTG bar image can well characterize different brightness case in natural image from black to white in the present invention, by using different removing-noise strength parameters to this GTG bar image denoising, the suitable strength denoising parameter under each brightness can be found, and the best removing-noise strength parameter obtained by the method for interpolation under all brightness, thus be applied to real image denoising, the denoising parameter of varying strength can be adopted the pixel of the different brightness of each in real image, obtain better denoising effect.

Fig. 3 adopts the design sketch after the final denoising of the inventive method to the image of Fig. 2 Noise, and adopt the inventive method to obtain the image denoising effect after denoising as can be seen from Figure 3 good, the edge details of image etc. obtain good reservation.

Claims (4)

1. the adaptive non-local mean image de-noising method of noise intensity, is characterized in that, specifically comprise the following steps:

Step 1: gather GTG bar image, inputted computing machine, GTG bar image is designated as z (i), and wherein i represents pixel, and z represents the brightness value of this pixel, and luminance areas different for GTG bar is designated as X _m;

Step 2: obtain the best removing-noise strength parameter under different brightness;

2.1, use non-local mean method under different removing-noise strength parameter, to carry out denoising to GTG bar image z (i), to obtain different brightness Y _mcorresponding best removing-noise strength parameter g _m, be designated as (g _m, Y _m), Y _mfor luminance area X _maverage brightness;

2.2, adopt linear difference method to obtain and be different from brightness Y _mbrightness P _ncorresponding best removing-noise strength parameter q _n;

Step 3: best denoising parameter corresponding to the different brightness obtained according to step 2 treats the noise intensity self-adaptive solution that denoising image carries out under different brightness.

2. the adaptive non-local mean image de-noising method of a kind of noise intensity according to claim 1, is characterized in that, the detailed process of described step 2.1 is:

2.1.1, adopt denoising formula to GTG bar image z (i) denoising:

By removing-noise strength parameter h from 1, change value successively from small to large, carry out denoising to GTG bar image z (i), obtain the image after a series of denoising, wherein, the value of removing-noise strength parameter h is: h ₁=1, h _i=10 (i-1), 2≤i≤101, i is integer, h _irepresent the value of i-th removing-noise strength parameter, denoising formula is as follows:

NLM(i)＝Σω(i,j)z(j)

$\mathrm{\ω}(i,j)=\frac{1}{C\left(i\right)}\exp (-\frac{{\left|\right|z\left(N_i\right)-z\left(N_j\right)\left|\right|}_2^2}{h^2})$

$C\left(i\right)=\underset{j}{\mathrm{\Σ}}\exp (-\frac{{\left|\right|z\left(N_i\right)-z\left(N_j\right)\left|\right|}_2^2}{h^2});$

Wherein, NLM (i) is the image after gray scale image z (i) being used the denoising of non-local mean method, and ω (i, j) represents the weight between pixel i and j, and meet 0≤ω (i, j)≤1 and j is the pixel in 21 × 21 regions centered by i; C (i) is normalized factor, N _irepresent the image block of 7 × 7 centered by pixel i, N _jrepresent the image block of 7 × 7 centered by pixel j;

2.1.2, GTG bar image z (i) each luminance area X is calculated _madopt different removing-noise strength parameter h _ithe Y-PSNR PSNR of a series of images obtained after denoising, chooses each luminance area X _mremoving-noise strength parameter h corresponding to the image that in a series of images obtained after denoising, Y-PSNR PSNR is maximum _ias corresponding bright region X _mbest removing-noise strength parameter, be designated as g _m, and calculate each luminance area X _maverage brightness Y _m, by each luminance area X _munder best removing-noise strength parameter and average brightness be expressed as (g _m, Y _m);

The computing formula of Y-PSNR PSNR is:

$\mathrm{PSNR}=10\log \left(\frac{{255}^2}{\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{(\mathrm{NLM}\left(i\right)-z\left(i\right))}^2}\right);$

Wherein, M represents luminance area X _msum of all pixels;

The average brightness Y of each luminance area _mcomputing formula is:

$Y_m=\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}z\left(i\right).$

3. the adaptive non-local mean image de-noising method of a kind of noise intensity according to claim 1, is characterized in that, the detailed process of described step 2.2 is:

If brightness P _nnot at (g _m, Y _m) among, then from (g _m, Y _m) among find adjacent two average brightness the most close with its numerical value, be designated as Y respectively _jand Y _j+1, wherein Y _jbe less than P _n, Y _j+1be greater than P _n, Y _jand Y _j+1at (g _m, Y _m) in corresponding removing-noise strength parameter be respectively g _jand g _j+1, then brightness P _ncorresponding best removing-noise strength parameter q _ntried to achieve by linear interpolation formulae discovery, linear interpolation formula is:

$q_n=\frac{|P_n-Y_j|}{|Y_{j+1}-Y_j|}\×g_j+\frac{|Y_{j+1}-P_n|}{|Y_{j+1}-Y_j|}\×g_{j+1}.$

4. the adaptive non-local mean image de-noising method of a kind of noise intensity according to claim 1, is characterized in that, the detailed process of described step 3 is:

3.1, by treating that the image of denoising is expressed as g (x, y), carry out natural logarithm conversion to it, transformation results is g ₁(x, y)=lng (x, y);

3.2, at log-domain to g ₁(x, y) adopts non-local mean method to carry out denoising, is specially:

To g ₁each pixel (x, y) of (x, y), adopts the removing-noise strength parameter h of the best that its original image g (x, y) brightness value is corresponding, according to denoising formula to g ₁(x, y) denoising, obtains the image after denoising, is designated as g ₂(x, y);

3.3, to g ₂(x, y) carries out exponential transform, obtains the image h (x, y) after final denoising, and result is h (x, y)=exp (g ₂(x, y)).