基于matlab图像修复算法仿真,对比插值,OMP以及KSVD三种算法

clc; clear; close all; warning off; addpath(genpath(pwd)); Case = 2; bb = 8; % block size K = 256; % number of atoms in the dictionary overlap = 3; % number of pixels between consecutive patches J = 5; % Number of iteration in KSVD img =imresize(imread('dat.png'),1); switch Case %% Interpolation case 1 Mask = double(~(img(:,:,1)==0)); for channel = 1:3 IMin0 = img(:,:,channel); IMin0= double(IMin0); out(:,:,channel) = Interpolation(IMin0,~Mask); end imshow(uint8(out)) %% OMP case 2 [N,M,dim]=size(img); img = double(img); NN = ceil((N-bb)/overlap) * overlap + bb; MM = ceil((M-bb)/overlap) * overlap + bb; img_new = 255 * zeros(NN,MM,dim); img_new(1:N,1:M,:) = img; %imshow(uint8(img_new)) %Compute mask and extracting its patches Mask = double(~(img_new(:,:,1)==0)); blkMask = overlap_im2col(Mask, bb, overlap); img_yuv = rgb2ycbcr(uint8(img_new)); img_inpaint_yuv = zeros(size(img_yuv)); % Interpolation CbCr Componet img_inpaint_yuv(:,:,2) = Interpolation(double(img_yuv(:,:,2)),~Mask); img_inpaint_yuv(:,:,3) = Interpolation(double(img_yuv(:,:,3)),~Mask); IMin0 = double(img_yuv(:,:,1)); % The dictionary DCT DCT=zeros(bb,sqrt(K)); for k=0:1:sqrt(K)-1, V=cos([0:1:bb-1]'*k*pi/sqrt(K)); if k>0 V=V-mean(V); end; DCT(:,k+1)=V/norm(V); end; DCT=kron(DCT,DCT); % Extracting the noisy image patches blkMatrixIm = overlap_im2col(IMin0, bb, overlap); sigma = 0.001; % 0.005 0.01 rc_min = 0.01; % rc_min: minimal residual correlation before stopping pursuit max_coeff = 10; % max_coeff: sparsity constraint for signal representation 10 % Inpainting the Patches (OMP) Coeff=OMP_Inpainting(DCT,blkMatrixIm.*blkMask,blkMask,sigma,rc_min,max_coeff); %256*146405 Coeff = full(Coeff); % Creating the output image imag_Y = overlap_col2im(DCT*Coeff, blkMask, bb, overlap, size(img_new)); imag_Y=max(min(imag_Y,255),0); img_inpaint_yuv(:,:,1) = imag_Y; img_inpaint_rgb = ycbcr2rgb(uint8(img_inpaint_yuv)); img_out = img_inpaint_rgb(1:N,1:M,:); imshow(uint8(img_out)) %% KSVD case 3 [N,M,dim]=size(img); img = double(img); NN = ceil((N-bb)/overlap) * overlap + bb; MM = ceil((M-bb)/overlap) * overlap + bb; img_new = 255 * zeros(NN,MM,dim); img_new(1:N,1:M,:) = img; %imshow(uint8(img_new)) %Compute mask and extracting its patches Mask = double(~(img_new(:,:,1)==0)); blkMask = overlap_im2col(Mask, bb, overlap); img_yuv = rgb2ycbcr(uint8(img_new)); img_inpaint_yuv = zeros(size(img_yuv)); % CbCr Componet Interpolation img_inpaint_yuv(:,:,2) = Interpolation(double(img_yuv(:,:,2)),~Mask); img_inpaint_yuv(:,:,3) = Interpolation(double(img_yuv(:,:,3)),~Mask); IMin0 = double(img_yuv(:,:,1)); % The dictionary DCT DCT=zeros(bb,sqrt(K)); for k=0:1:sqrt(K)-1, V=cos([0:1:bb-1]'*k*pi/sqrt(K)); if k>0 V=V-mean(V); end; DCT(:,k+1)=V/norm(V); end; DCT=kron(DCT,DCT); % Extracting the noisy image patches blkMatrixIm = overlap_im2col(IMin0, bb, overlap); sigma = 0.01; % 0.005 0.01 rc_min = 0.01; % rc_min: minimal residual correlation before stopping pursuit max_coeff = 10; % max_coeff: sparsity constraint for signal representation 10 % Inpainting the Patches (K-SVD) [Dict Coeff]=KSVD_Inpainting(DCT,blkMatrixIm,blkMask,sigma,rc_min,max_coeff,J); %DCT % Creating the output image imag_Y = overlap_col2im(Dict*Coeff, blkMask, bb, overlap, size(img_new)); imag_Y=max(min(imag_Y,255),0); img_inpaint_yuv(:,:,1) = imag_Y; img_inpaint_rgb = ycbcr2rgb(uint8(img_inpaint_yuv)); img_out = img_inpaint_rgb(1:N,1:M,:); imshow(uint8(img_out)) end