Hsu.zip_Color Filter Array_nature_signature

function [Ppost_final, trials, alpha] = classify_single (image) % Start EM algorithm [x, y, z] = size(image); for chan=1:3 % Initialization alpha = 1/8 * ones(3, 3); alpha(2, 2) = 0; sigma = 0.0075; %P0 = 1/ 256; P0 = 1/((sigma *(2*pi)^0.5)); img = double(image(:,:,chan)) ./ 255 ; N = 1; % want 3x3 square where middle square is 0 error = 5; trials = 0; epsilon = sigma / 100; while (error > .01) & (trials < 100) % Expectation est = conv2(img, alpha, 'valid'); res = img(2:end-1, 2:end-1) - est; %epsilon = sigma / (100); %Pcond = 1/(sigma *(2*pi)^0.5) * (exp(-(res).^2/(2*sigma^2))) * 2 * epsilon; Pcond = 1/(sigma *(2*pi)^0.5) * (exp(-(res).^2/(2*sigma^2))); %P0 = 1/(sigma *(2*pi)^0.5) /255; Ppost = Pcond ./ (Pcond + P0); % Generate f(x+u, y+v) matrices first for v=-N:N for u=-N:N eval(['img' num2str(u+N) num2str(v+N) '=img(2+' ... num2str(u) ':end-1+' num2str(u) ', 2+' num2str(v) ':end-1+' num2str(v) ');']); end end count = 1; % A and B matrices Afinal = zeros(1, ((2*N)+1)^2-1); for t=-N:N for s=-N:N BaseMatrix = Ppost .* img(2+s:end-1+s, 2+t:end-1+t); Bmatrix = Ppost .* img(2+s:end-1+s, 2+t:end-1+t) .* img(2:end-1, 2:end-1); B(count) = sum(sum(Bmatrix)); count = count+1; Anew = [sum(sum(BaseMatrix .* img00)), ... sum(sum(BaseMatrix .* img10)), ... sum(sum(BaseMatrix .* img20)), ... sum(sum(BaseMatrix .* img01)), ... sum(sum(BaseMatrix .* img21)), ... sum(sum(BaseMatrix .* img02)), ... sum(sum(BaseMatrix .* img12)), ... sum(sum(BaseMatrix .* img22))]; Afinal = [Afinal ; Anew]; end end Afinal = Afinal (2: end, :); alphaold = alpha; B = [B(1:4), B(6:end)]; Afinal = [Afinal(1:4, :) ; Afinal(6:end,:)]; alpha = linsolve (Afinal, B'); alpha = [alpha(1:4); 0; alpha(5:end)]; alpha = [alpha(1:3), alpha(4:6), alpha(7:9)]; sigma = ((sum(sum(Ppost .* res.^2)))/(sum(sum(Ppost)))).^.5; error = sum(sum(abs(alphaold-alpha))); trials = trials + 1; end % End entire algorithm alpha_final( :,:, chan) = alpha; Ppost_final(:,:,chan) = Ppost; end %} error; alpha; trials; end