matlab-(含教程)基于量子自适应变换QAB算法的图像去噪matlab仿真

function [I_result] = image_denoising_QAB(I,J,Ms,pds,sg) %size of image [M,M1] = size(I); % assume M = M2 MM = M^2; N = 64; NN = N^2; P = 2*M/N; % store initial data I_old = I; J_old = J; saut = 12; %Moving the threshold (must be at least 1, this is the slowest case) % Threshold parameter Vs = linspace(7,11,Ms); % for reconstruction Vs = 2 .^Vs; disp('Start the search of wave function') J_new = zeros(M,M); %creat space for the big image o = 0; cmpt = zeros(M,M); %count the overlapping for i = 0:(P-2) for j = 0:(P-2) % These variables store the values associated with the maximums. V_ms = 0.01; RSB_ms = 0; % maximum SNR for adaptive transformation seuil_ms = 1; % Devide the image in to small parts J_part = J((i*N/2+1):(i*N/2+N),(j*N/2+1):(j*N/2+N)); I_part = I((i*N/2+1):(i*N/2+N),(j*N/2+1):(j*N/2+N)); % Normalize small part J_part_max = max(max(J_part)); I_part = I_part/J_part_max; J_part = J_part/J_part_max; pI_part = sum(sum(I_part .^2)) / NN; % The noisy image is blurred % Creation of a Gaussian [x,y] = meshgrid((-N/2):(N/2 - 1)); z = 1 / (sqrt(2 * pi * sg))^2 * exp(-(x .^2 + y .^2) / (2 * sg)); % Convolution product gaussF = fft2(ifftshift(z)); JF = fft2(J_part); n_J = real(ifft2(gaussF .* JF)); % new images Jns = zeros(N,N); %Calculation of the eigenvectors then projection in the base J_col = reshape(J_part,NN,1); [psi,psi_col,E] = f_ondes2D(n_J,pds); alp = linsolve(psi_col,J_col); % Reconstruction des images for k = 1:Ms %Loop on the slope parameter v = Vs(k); max_atteint = 0; RSB_s = 0; l = 1; while (l <= NN) * (max_atteint < 20) %Loop on the threshold position, stops when one reaches a local maximum (global?) For the threshold parameter % threshold x = (1:(NN)) - l + 2; taux = heavi(x,v); %Re-construction n_I = zeros(N,N); for t = 1:(NN) n_I = n_I + psi(:,:,t) * taux(t) * alp(t); end % Calculation of SNR n_B = n_I - I_part; pnB = sum(sum(n_B .^2)) / NN; RSB_n = 10 * log10(pI_part / pnB); % Test if the SNR increases max_atteint = (max_atteint + 1) * (RSB_s > RSB_n); RSB_s = RSB_s * (RSB_s > RSB_n) + RSB_n * (RSB_s <= RSB_n); V_ms = V_ms * (RSB_ms >= RSB_n) + v * (RSB_ms < RSB_n); seuil_ms = seuil_ms * (RSB_ms >= RSB_n) + l * (RSB_ms < RSB_n); Jns = Jns * (RSB_ms >= RSB_n) + n_I * (RSB_ms < RSB_n); RSB_ms = RSB_ms * (RSB_ms >= RSB_n) + RSB_n * (RSB_ms < RSB_n); l = l + saut; end end %store the small part into the big image J_new((i*N/2+1):(i*N/2+N),(j*N/2+1):(j*N/2+N)) = Jns * J_part_max + J_new((i*N/2+1):(i*N/2+N),(j*N/2+1):(j*N/2+N)); cmpt((i*N/2+1):(i*N/2+N),(j*N/2+1):(j*N/2+N)) = cmpt((i*N/2+1):(i*N/2+N),(j*N/2+1):(j*N/2+N)) + 1; o = o+1 end end J_new = J_new./cmpt; %remove the overlapping I_result = J_new; end