i need help to understand the steps of this code
2 views (last 30 days)
Show older comments
hi everyone please can someone explain to me the steps of this program ?
its a coherence filter program
thanks for helping
%%%
function myCEDnew( )
myalpha = 0.001; sigma=0.7; T = 15; rho = 4; C= 1;
im = imread('../images/2.png');
[numrow, numcol] = size(im);
imorig = im;
stepT=0.15;
t = 0;
im=double(im);
while (t < (T-0.001))
t = t + stepT;
%% 1 gaussian K_sigma
limitX=-ceil(2*sigma):ceil(2*sigma);
kSigma = exp(-(limitX.^2/(2*sigma^2)));
kSigma = kSigma/sum(kSigma(:));
usigma=imfilter(imfilter(im,(kSigma'), 'same' ,'replicate'),kSigma, 'same' ,'replicate');
%% Gradient
[uy,ux]=gradient(usigma);
%% 3 gaussian K_rho
limitXJ=-ceil(3*rho):ceil(3*rho);
kSigmaJ = exp(-(limitXJ.^2/(2*rho^2)));
kSigmaJ = kSigmaJ/sum(kSigmaJ(:));
Jxx = imfilter(imfilter((ux.^2),(kSigmaJ'), 'same' ,'replicate'),kSigmaJ, 'same' ,'replicate');
Jxy = imfilter(imfilter((ux.*uy),(kSigmaJ'), 'same' ,'replicate'),kSigmaJ, 'same' ,'replicate');
Jyy = imfilter(imfilter((uy.^2),(kSigmaJ'), 'same' ,'replicate'),kSigmaJ, 'same' ,'replicate');
%% Principal axis transformation
% Eigenvectors of J, v1 and v2
v2x = zeros(numrow, numcol);
v2y = zeros(numrow, numcol);
lambda1 = zeros(numrow, numcol);
lambda2 = zeros(numrow, numcol);
for i=1:numrow
for j=1:numcol
pixel = [Jxx(i,j), Jxy(i,j); Jxy(i,j), Jyy(i,j)];
[pixelV, pixelD] = eig(pixel);
v2x(i,j) = pixelV(1,2);
v2y(i,j) = pixelV(2,2);
lambda1(i,j) = pixelD(1,1);
lambda2(i,j) = pixelD(2,2);
if((v2x(i,j)^2)+(v2y(i,j)^2)==0)
abcd=0;
else
v2x(i,j) = v2x(i,j)/(sqrt((v2x(i,j)^2)+(v2y(i,j)^2)));
v2y(i,j) = v2y(i,j)/(sqrt((v2x(i,j)^2)+(v2y(i,j)^2)));
end;
end;
end;
v1x = -v2y;
v1y = v2x;
%% Calculation of diffusion matrix
di=(lambda1-lambda2);
lambda1 = myalpha + (1 - myalpha)*exp(-C./(di).^(2));
lambda2 = myalpha;
Dxx = lambda1.*v1x.^2 + lambda2.*v2x.^2;
Dxy = lambda1.*v1x.*v1y + lambda2.*v2x.*v2y;
Dyy = lambda1.*v1y.^2 + lambda2.*v2y.^2;
%% Non negativity discretization scheme referred from http://citeseer.ist.psu.edu/viewdoc/download;jsessionid=16CCEAD1A72E6A1CC960DF99795D62B7?doi=10.1.1.21.632&rep=rep1&type=pdf
im=non_negativity_discretization(im,Dxx,Dxy,Dyy,stepT);
end;
%% output
figure(1);
subplot(1, 2, 1);
imagesc(imorig);
title('Original image');
colormap('Gray');
daspect ([1 1 1]);
subplot(1, 2, 2);
imagesc(im);
title('Coherence Enhancing Diffusion Filtering');
colormap('Gray');
daspect ([1 1 1]);
figure(2);
subplot(1, 2, 1);
imagesc(imorig);
title('Original image');
colormap('Gray');
daspect ([1 1 1]);
subplot(1, 2, 2);
imagesc(atan2(double(uy), double(ux)));
title('Orientation of smooth gradient');
colormap('Gray');
daspect ([1 1 1]);
imwrite((uint8(im)), '../images/2CED.png');
end
function im=non_negativity_discretization(im,Dxx,Dxy,Dyy,stepT)
% Make positive and negative indices
[numrow,numcol] = size(im);
px = [2:numrow,numrow]; nx = [1,1:numrow-1];
py = [2:numcol,numcol]; ny = [1,1:numcol-1];
% In literature a,b and c are used as variables
a=Dxx;b=Dxy;c=Dyy;
% Stencil Weights
wbR1 = (0.25)*((abs(b(nx, py))-b(nx,py)) + (abs(b)-b));
wtM2 = (0.5)*( (c(:, py)+c) -(abs(b(:,py))+abs(b)));
wbL3 = (0.25)*((abs(b(px, py))+b(px,py)) + (abs(b)+b));
wmR4 = (0.5)*( (a(nx,: )+a) -(abs(b(nx,:))+abs(b)));
wmL6 = (0.5)*( (a(px,: )+a) -(abs(b(px,:))+abs(b)));
wtR7 = (0.25)*((abs(b(nx, ny))+b(nx,ny)) + (abs(b)+b));
wmB8 = (0.5)*( (c(:, ny)+c) -(abs(b(:,ny))+abs(b)));
wtL9 = (0.25)*((abs(b(px, ny))-b(px,ny)) + (abs(b)-b));
im= im+stepT*(wbR1.*(im(nx,py) -im(:,:))+wtM2.*(im(:, py) -im(:,:))+wbL3.*(im(px,py) -im(:,:))+wmR4.*(im(nx,:) -im(:,:))+ ...
wmL6.*(im(px,:) -im(:,:))+ wtR7.*(im(nx,ny) -im(:,:))+ wmB8.*(im(:, ny) -im(:,:))+ wtL9.*(im(px,ny) -im(:,:)));
end
1 Comment
Image Analyst
on 29 May 2019
I doubt any of us can/would, but the authors probably would. Have you asked them?
Answers (0)
See Also
Categories
Find more on Geometric Transformation and Image Registration in Help Center and File Exchange
Products
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!