Till Halbach.
Robust Still Image Compression for Mobile and Wireless Communications.
Diploma thesis,
NTNU, Trondheim (Norway),
December 1998

Abstract
In this thesis, a robust and efficient still image compression/ decompression system based on a fixed-length bit stream syntax is developed.
In the first part of the thesis, the current state of the standardization efforts of the JPEG2000 project is reviewed. The current release of the Verification Model (VM), i.e., 2.1,  is investigated with regard to a bit stream syntax which is robust to channel errors. The VM2.1 encoding system consists of a subband decomposition, i.e., a discrete wavelet transform, a trellis coded quantizer and a following binary arithmetic bit plane encoder with causal contexts which provides variable-length code words to be transmitted over error-prone channels.
Fixed-length coding is then, in the second part of the thesis, considered to enable the decoder to re-establish synchroniza-tion which is usually lost when transmitting variable-length codes over noisy channels. For a given channel capacity, the objective is to minimize the mean-squared error between the wavelet samples to be quantized and the reconstructed samples. Nonuniform scalar Lloyd-Max quantizers are then employed,  taking both the source statistics and the rate constraint into account. Also, scalar quantization (SQ) requires an optimal bit allocation procedure which is derived with respect to the statistics of the wavelet samples and the preceeding subband decomposition. For binary symmetric channels and burst error channels, it is demonstrated that SQ yields significant improvements in PSNR, whereas in the error-free case image quality degradations have to be accepted.
In the last part of the thesis, two codec structures for progressive coding using nonuniform scalar quantizers are investigated. First, a look at the VM with included SQ, where transform-based hierarchical coding is enabled, is taken. There, it is possible to decode progressively, whereas it is found that a hierarchical tree structure of scalar quantizers is not efficient for this purpose.

Keywords
Image processing, data compression, robust communications, JPEG2000 standardization, error resilience, scalar quantization.