How correlation helps in compression

Question

Hy, I would like to know how correlation between pixels in image helps in compression? Also, why I would like to reduce correlation between pixels to get better compression (I found that information in one literature, but I don't understand it)? My last question is: If correlation is R=0.9, how this information can help in compressing?

Thanks.

Answer 1

I'll make an example. Let's say that every pixel is very much correlated with the pixel above it. Instead of compressing the pixel values directly we now compress bottomPixel - topPixel (the delta of the two).

This delta will be very small because of the correlation. I can now use a variable-length encoding like arithmetic or huffman to assign less bits to small deltas and more bits to high deltas. I'll save bits that way because small deltas occur a lot more often.

This example generalizes to lots of cases and models (audio being one of them).