Parse Forests with Huffman-Tree-Based Modes for One-Symbol-Delay Encodable VF Codes

We present an efficient construction of variable-to-fixed-length (VF) codes under conditions where encoding delay is allowed. We introduce a generalized form of parsetree sets representing a class of codes broader than the almost instantaneous VF (AIVF) codes, which are the conventional VF codes that allow one symbol of an encoding delay. Using Huffman trees, we can optimally construct the codes and guarantee that the compression efficiency is at least compatible with Huffman codes. Numerical evaluations show that we can make efficient VF codes that outperform Huffman codes even for input sources that are disadvantageous to conventional VF codes.