How to find the variance of the coefficients in each subband using Bayes Shrink method.

I'm performing compression on an image using bayes shrink method. so i first added gaussian noise to my image and applied 2D discrete wavelet transform.

So I found the variance of the HH subband using the following Matlab code.

sigma=median(abs(cH))/0.6475;

But I cant square this value because it says it has to be a square matrix.

Next in order to find variances in each subband I'm stuck. I have to find a summation of all coefficients in the subband. But I'm not quite sure how to do this for greater than 1 level of decompostion. Please help with a sample code.

Answer 1

[Walter Roberson]

You probably want

sigma.^2

You probably tried

sigma^2

which is sigma matrix-multiplied by sigma

Answer 2

[Audrey]

Thanks. Any idea how I can find thresholds for each sub-band for higher levels of decomposition?

[Walter Roberson]

Sorry, I have not studied wavelets.

Answer 3

[bushra khan]

N=1024;
L= 4;
t=0:1:N-1;
signal=wnoise(2,10);
figure;
snr=2;
y = awgn(signal,snr,'measured');
noisy = signal + y; %noisy signal
plot(t,noisy);
title('noisy signal');
[c,l] = wavedec(noisy,L,'db1');   %wavelet
A1 = appcoef(c,l,'db1');
d1 = detcoef(c,l,1);
sig=median(abs(d1))/0.6745;
ct=[];
for r = 1:L 
   cdr = detcoef(c,l,r); %detailed coefficients
   sigy=var(cdr);
   sigx=sigy-sig^2;
   z=sqrt(sigx);
   T=(sig^2)/sigx;
   cdr=cdr-T;   %soft thresholding
   cdr(cdr<=T)=0;
   ct=[cdr ct];
end
ct=[A1 ct];
X = waverec(ct,l,'db1');
figure;
%hold on
plot(X);
title('reconstruction using hard thresholding');