robust_scale_filter.rar_scale

function [out]=robust_scale_filter(img_bruit,image_no_bruit,fonc_diff,nbre_iter,seuil_contour,dt) % Peronna-Malick Anisotropic Diffusion % I:input gray or color image %% img_bruit:image contenant du bruit en niveau de gris ou % couleur % fonc_diff:(fonction de diffusion ou d'arrêt contour) % ='lin':linear diffusion, g=1 % ='pm1':perona_malik, g=1/(1+(x/K)^2) % ='pm2':perona-malik, g=exp(-(x/K)^2) % ='tky':Tukey's biweight, g= / (1-(x/K).^2).^2 ,|x|<=K % \ 0 ,otherwise % method='pm1 ou pm2':Peronna et Malick, ut=div(g(|Du|)*Du) % nbre_iter:nombre d'itération défaut= 100 % seuil_contour: paramètre seuil contour [row,col,nchannel]=size(img_bruit); i=1; psnr_sup=1; psnr_inf=0; % val_mae=1; %while val_mae>seuil_contour% while round(psnr_sup,3)>round(psnr_inf,3)%for i=1:nbre_iter disp('RSF') i=i+1 %% %calcul des valeurs du gradient dans les quatre directions (N,S,E,O) % N % O C E % S grad_N=[img_bruit(1,:,:);img_bruit(1:row-1,:,:)]-img_bruit; %N-C grad_S=[img_bruit(2:row,:,:);img_bruit(row,:,:)]-img_bruit; %S-C grad_E=[img_bruit(:,2:col,:) img_bruit(:,col,:)]-img_bruit; %E-C grad_O=[img_bruit(:,1,:) img_bruit(:,1:col-1,:)]-img_bruit; %O-C %Compute the coefficient (k,a) of normalization g() if strcmp(fonc_diff,'lin') k_N=1; k_S=1; k_E=1; k_O=1; elseif strcmp(fonc_diff,'pm1') k_N=2; k_S=2; k_E=2; k_O=2; a=1;%amplitude normalization elseif strcmp(fonc_diff,'pm2') k_N=1.4826*mad(grad_N(:),1)*(2^0.5); k_S=1.4826*mad(grad_S(:),1)*(2^0.5); k_E=1.4826*mad(grad_E(:),1)*(2^0.5); k_O=1.4826*mad(grad_O(:),1)*(2^0.5); a=1/(2*exp(-0.5)); elseif strcmp(fonc_diff,'tky') k_N=1.4826*mad(grad_N(:),1)*(5^0.5); k_S=1.4826*mad(grad_S(:),1)*(5^0.5); k_E=1.4826*mad(grad_E(:),1)*(5^0.5); k_O=1.4826*mad(grad_O(:),1)*(5^0.5); a=25/32; end %% calcul de la fonction de diffusion if strcmp(fonc_diff,'lin')%isotropic diffusion (gaussian) g_N=k_N;g_S=k_S;g_E=k_E;g_O=k_O; elseif strcmp(fonc_diff,'pm1')%anisotropic diffusion(PM1) g_N=1./sqrt(1+(grad_N/k_N).^2).*a; g_S=1./sqrt(1+(grad_S/k_S).^2).*a; g_E=1./sqrt(1+(grad_E/k_E).^2).*a; g_O=1./sqrt(1+(grad_O/k_O).^2).*a; elseif strcmp(fonc_diff,'pm2')%anisotropic diffusion (PM2) g_N=1./(1+(grad_N/k_N).^2).*a; g_S=1./(1+(grad_S/k_S).^2).*a; g_E=1./(1+(grad_E/k_E).^2).*a; g_O=1./(1+(grad_O/k_O).^2).*a; elseif strcmp(fonc_diff,'tky')%anisotropic diffusion (Tukey) g_N=zeros(row,col,nchannel);g_S=g_N;g_E=g_N;g_O=g_N; indexn=find(abs(grad_N)<=k_N); indexs=find(abs(grad_S)<=k_S); indexe=find(abs(grad_E)<=k_E); indexw=find(abs(grad_O)<=k_O); g_N(indexn)=(1-(grad_N(indexn)/k_N).^2).^2*a; g_S(indexs)=(1-(grad_S(indexs)/k_S).^2).^2*a; g_E(indexe)=(1-(grad_E(indexe)/k_E).^2).^2*a; g_O(indexw)=(1-(grad_O(indexw)/k_O).^2).^2*a; end %% img_bruit_n1=img_bruit; %équation de diffusion img_bruit=img_bruit_n1+dt*((g_N.*grad_N+g_S.*grad_S+g_E.*grad_E+g_O.*grad_O)./(g_N+g_S+g_E+g_O)); TABSCISSE(i)=i; TPSNR(i)=psnr(img_bruit,image_no_bruit); TMAE(i) =mean2(abs(img_bruit-img_bruit_n1)); TSSIM(i)=ssim(img_bruit,image_no_bruit); %val_mae=MAE(i) psnr_sup=TPSNR(i) psnr_inf=TPSNR(i-1) end out.NBR_ITER=max(TABSCISSE); if TPSNR(out.NBR_ITER)>TPSNR(out.NBR_ITER-1) out.PSNR=TPSNR(out.NBR_ITER); out.SSIM=TSSIM(out.NBR_ITER); out.NBR_ITER=out.NBR_ITER; else out.PSNR=TPSNR(out.NBR_ITER-1); out.SSIM=TSSIM(out.NBR_ITER-1); out.NBR_ITER=out.NBR_ITER; end out.img_filtre=img_bruit; end