A finite-state multiresolution approach to image compression using pruned nested tree-structured vector quantization

An algorithm is described for constructing a finite-state compression code that is both progressive and multiresolution. The codec consists of nested levels of tree-structured vector quantizers (TSVQs) where the codebook for each level of the nested TSVQs is constructed from the terminal leaves of the TSVQ from the previous level. The first level of the TSVQ represents a finite-state next state function. The codeword dimension at the subsequent levels is greater than or equal to those of the previous levels. This property allows the codec to produce a multiresolution output in a progressive manner. Pruning is performed on the nested TSVQs to achieve the bit allocation across the levels. The resulting pruned TSVQ decoder operates entirely by successive table lookups, with no arithmetic computation. Furthermore, it provides superior performance to ordinary pruned TSVQ at low bit rates.<>