A new vector quantization for wavelet image coding

This paper proposes a new scheme for image coding using the wavelet transform. We decompose the original image into seven subbands with a two-level tree-structured filter bank. Vector quantization is applied to the coefficients of the high-frequency subbands, and 2-D DPCM to the lowest-frequency subband. A vector is formed by unifying n significant coefficients whose amplitude is larger than a preassigned threshold. Then, in order to compress the position information of those significant coefficients, we decide whether it is significant or not hierarchically by each coefficient block instead of by each coefficient. This scheme gives coding performance superior to conventional quantization of vectors unifying both significant and insignificant coefficients. (C) 1997 Scripta Technica, Inc.