SNR: signal to noise ratio i.e., -20*log10
(L2_norm(signal)/L2_norm(noise))
BSNR: blurred signal to noise ratio i.e., -20*log10 (Var(
conv(signal,blur) )/L2_norm(noise)), where conv denotes convolution;
and Var denotes variance, hence, conv(signal,blur) is made zero-mean,
and then the L2_norm is taken to obtain the Var(conv(signal,blur)).
ISNR: improvement in signal to noise ratio i.e., -20*log10
(L2_norm(signal - observation)/L2_norm(signal - estimate))
The observation is shifted to get the lowest ISNR. For example, with a
9x9 box-car blur, the observation is shifted by (4,4) towards the top
left hand corner before calculating the BSNR terms.
<\FONT>