Codebook generation and search algorithm for vector quantization using arbitrary hyperplanes

An arbitrary hyperplane approach for both codebook generation and search for vector quantization (VQ) is introduced. It is a divisive algorithm and aims at minimizing the sum-of-squared-errors. The relation to tree-structured vector quantization is discussed, and the superiority of the proposed algorithm is demonstrated through a comparison with the LBG (generalized Lloyd) algorithm for a 16-dimension VQ compression of still pictures.<
>