TABLE OF CONTENTS Source encoding is a natural constraint, due to the need to adapt to transport support or information storage. The goal is then to translate from one symbol to the other so as to make the best possible use of the resource (transmission channel or storage medium), assumed to be error-free. Moreover, as will become clear for images, various kinds of redundancy make it possible to associate coding with specific compression techniques.
In light of what we have seen previously, we can understand that the goal of source coding is to obtain another source of maximum entropy. Here, we look for a method yielding the optimal code regardless of the source symbol probabilities. As we have seen, for an optimal code the shortest words should be associated with the most probable source symbols. It follows that, given codewords of known lengths, encoding consists of sorting these according to increasing lengths so as to match the longest words with the source symbols that occur least. This is entropy coding. It should be recalled that only codewords enjoying the prefix property can be taken into account (they are sometimes also called prefix-free); thus, the source alphabet must be used to build codewords of increasing length obeying this property.
These are the bases on which various encoding algorithms have been developed. We first illustrate the principle, through the presentation of the fundamental Huffman coding. Then we address the general issue of compression, for which a large variety of...
