An empirical evaluation of coding methods for multi-symbol alphabets

The authors examine the resource requirements and compression efficiency of the coding phase, concentrating on applications with medium and large alphabets. When semi-static two-pass encoding can be used, Huffman coding is two to four times faster than arithmetic coding, and sometimes results in superior compression. When an adaptive coder is required the difference in speed is smaller, but Gallager's implementation of dynamic Huffman coding is still faster than arithmetic coding in most situations. The compression loss through the use of Huffman codes is negligible in all but extreme circumstances. Where very high speed is necessary splay coding is also worth considering, although it yields poorer compression.>