Wavelet-based progressive image and video coding using trellis-coded space-frequency quantization

Abstract

Images and video are a very challenging topic in multimedia applications. Recently, with the explosion of the Internet, multiresolution source codes are necessary to better suit to the heterogeneity of the networks. This thesis first proposes a new wavelet-based image coding algorithm that takes benefit of a rate-distortion optimization truncation scheme at the encoder as well as an embedded bit stream at the decoder. The rate-distortion optimization is performed via Space-Frequency Quantization (SFQ) in which a zerotree pruning process is used to zero out insignificant wavelet coefficients. In this new SFQ algorithm, wavelet trees are mapped only one time via C structures to achieve a simpler manipulation of wavelet coefficients as well as a fast pruning process. Arithmetic coding is used after uniform quantization of the remaining wavelet coefficients to increase the compression ratio. Although the decoder is only optimal for rates fixed during the encoding, the proposed scheme achieves performances competitive with results published so far in the literature. To further compress wavelet coefficients, uniform quantization is replaced by Trellis-Coded Quantization to generate a progressive image coder using Trellis-Coded SFQ. The basic idea of the Trellis-Coded Space-Frequency Quantization (TCSFQ) is to take benefit from both SFQ and TCQ schemes: TCQ is only applied on a subset of wavelet coefficients while the other coefficients are discarded. For progressive transmission, we use an approximation to estimate Trellis-Coded Quantization indices partially transmitted when decoding at lower rates than the encoding one. Then, a progressive rate-distortion three-dimensional (3-D) wavelet video coder is presented by extending the concept of SFQ from 2-D to 3-D. A memory-constrained lifting-based 3-D wavelet transform is first deployed to process one part of the sequence at a time continuously, thus eliminating the boundary effects over groups of pictures. After 3-D SFQ, in which a rate-distortion based spatial-temporal tree-pruning process is used in conjunction with uniform quantization of wavelet coefficients, an efficient 3-D entropy coder is applied. Finally, a progressive 3-D TCSFQ coder is designed by extending TCSFQ to 3-D. Both 3-D SFQ and 3-D TCSFQ video coders outperform MPEG-4 for most sequences at the same bit rate.