Although SNR is a widely used measure of distortion, it often does not correlate with perceptual quality. In order to see this we conduct the following experiment. Using one of the images provided, generate two "reconstructed" images. For one of the reconstructions add a value of 10 to each pixel. For the other reconstruction, randomly add either -f-10 or -10 to each pixel.
(a) What is the SNR for each of the reconstructions? Do the relative values reflect the difference in the perceptual quality?
(b) Devise a mathematical measure that will better reflect the difference in perceptual quality for this particular case.