Lossless Image Compression – General Overview cmpe 263 Winter 2001

Lossless image compression depends entirely on statistical techniques, i.e., compression depends upon coding each outcome i of a statistical event E according to its own relative frequency p(i). The self-information l(i), measured in bits, of an outcome i is log (1/p(i)) (also written -log p(i)) where p(i) is the relative frequency. Thus, the coding part of the compression problem must, for 100% coding efficiency, increase the code-length by l(i) bits for each outcome i. The point to be made here, is that the encoding part can only worsen the achievable compression of a particular approach.