Method to 'clean' an image using hard-thresholding in the steerable wavelet domain (4 scales 8 orientations) The lambda parameter has been fitted using 100 images for variances between [200 - 400], it will not be optimal for different variances USE: IMM=hard_steerable(Im2,Var) Var is the noise variance
0001 % Method to 'clean' an image using hard-thresholding in the 0002 % steerable wavelet domain (4 scales 8 orientations) 0003 % 0004 % The lambda parameter has been fitted using 100 images for 0005 % variances between [200 - 400], it will not be optimal for different variances 0006 % 0007 % USE: IMM=hard_steerable(Im2,Var) 0008 % Var is the noise variance 0009 0010 function IMM=hard_steerable(Im2,Var) 0011 res= 0.7+sqrt(Var)*0.018; 0012 [pyr,ind]=buildsfpyr(Im2,4,7); 0013 pyr2=abs(pyr); 0014 pyr_sg=sign(pyr); 0015 load indices_epsilon_4esc_8ori 0016 indices=indices*sqrt(Var)/sqrt(400); 0017 res=res*indices; 0018 P2=zeros(size(pyr2)); 0019 for k=1:length(pyr2)-255 0020 if pyr2(k)>res(k) 0021 P2(k)=pyr2(k)-res(k); 0022 else 0023 P2(k)=0; 0024 end 0025 end 0026 P2(1:256*256-1)=zeros; 0027 P2(end-255:end)=pyr(end-255:end); 0028 PYR3=P2.*pyr_sg; 0029 0030 IMM=reconsfpyr(PYR3,ind);