Successive approximation trellis-coded vector quantization for embedded coders

Abstract

Embedded bit-plane-based and Trellis Coded Quantization-based (TCQ) methods are among the state-of-the-art for wavelet image coding. In the JPEG 2000 Verification Model, these two methods have been successfully combined in an Embedded TCQ Wavelet Coder. In it, TCQ is first applied to the coefficients and then its indexes are bit-plane encoded. In this paper, we propose a different philosophy of combining the aforementioned methods. Instead of encoding the TCQ indexes by bit-planes, we would apply TCQ to each bit-plane. However, since in scalar successive approximation processes, at every pass just a symbol from a two-valued source is transmitted (e.g. (+/spl delta/, -/spl delta/)), TCQ is not suitable for such scenario. We have circumvented this problem by using vector bit-planes. Each vector bit-plane can then be encoded using Trellis Coded Vector Quantization (TCVQ). Lattice-based codebooks were investigated and modified in order to yield efficient TCVQ schemes. Simulations have shown that the proposed method outperforms the Embedded Wavelet Vector Bit-plane encoder (SA-W-VQ). This indicates that such approach has the potential of increasing the coding performance of bit-plane-based methods.