Chapter 1 Huffman Coding

Codes may be characterized by how general they are with respect to the distribution of symbols they are meant to code. Universal coding techniques assume only a nonincreasing distribution. Golomb coding assumes a geometric distribution [1]. Fiala and Greene’s (start, step, stop) codes assume a piecewise uniform distribution function with each part distributed exponentially [2]. Huffman codes are more general because they assume nothing particular of the distribution; only that all probabilities are non-zero. This generality makes them suitable not only for certain classes of distributions, but for all distributions, including those where there is no obvious relation between the symbol number and its probability, as is the case with text. In text, letters go from ’a’ to ’z’ and are usually mapped onto a contiguous range, such as, for example, from 0 to 25 or from 97 to 122, as in ASCII, but there is no direct relation between the symbol’s number and its frequency rank.