© 1998 by British Computer Society
On the Data Expansion of the Huffman Compression Algorithm
1 MIT Laboratory for Computer Science, 545 Technology Square, Cambridge, MA 02139, USA Email: robdep{at}lcs.mit.edu, 2 Dipartimento di Informatica ed Applicazioni, Universitá di Salerno, 84081 Baronissi (SA), Italy
While compressing a file with a Huffman code, it is possible that the size of the file grows temporarily. This happens when the source letters with low frequencies (to which long codewords are assigned) are encoded first. The maximum data expansion is the average growth in bits per source letter resulting from the encoding of a source letter with a long codeword. It is a measure of the worst case temporary growth of the file. In this paper we study the maximum data expansion of Huffman codes. We provide some new properties of the maximum data expansion
of Huffman codes and using these properties we prove that
< 1.256.
Received January 27, 1998. revised May 8, 1998.