Code and parse trees for lossless source encoding

Efficient implementation of the generalized Tunstall code generation algorithm

“…Analyses of the performance of Tunstall's technique are prevalent in the literature [4], [5], but perhaps the most obvious advantage to Tunstall codes is that of being randomly accessible [5]: Each output block can be decoded without having to decode any prior block. This not only aids in randomly accessing portions of the compression sequence, but also in synchronization: Because the size of output blocks is fixed, simple symbol errors do not propagate beyond the set of input symbols a given output block represents.…”

New bounds on D-ary optimal codes

“…Tighter bounds can be obtained if we take into account the distribution. As explained, most of the huge amount of work on bounding the redundancy of D-ary optimal codes applies to the case D = 2 [1]. The only bounds we could find for general D are given in [13,4,5].…”

“…This is useful to predict the compression ratios that can be achieved by optimal codes on specific sources or for families of parametric distributions. Although a lot of work has been carried out on improving upper and lower bounds on the redundancy, most of it focuses on the case D = 2 (see an excellent survey by Abrahams [1], especially Section 2.2).…”

On improving Tunstall codes

“…Tunstall's procedure has been extensively analyzed [1], and the assumption of independent character appearance has been extended to sources with memory [15,17]. Our approach is a more practical one: instead of trying to model the text and choosing the dictionary based on the expected probabilities of the strings as induced by the model, we deal directly with the substrings that actually appear in the text and which can be processed by means of the text's suffix tree.…”