| One of the most significant challenges today in image coding and transmission is the need to handle errors in noisy channels. Since all communication channels have limited bandwidth, compression techniques are commonly adopted to reduce the redundancy within the digital information to be transmitted. Consequently, a strong data dependency in the coded data has been created. When some bits are corrupted with errors in the course of transmission, the compressed data may not be decoded correctly at the receiving end. Unfortunately, communication channels often suffer from various impairment, such as packet loss caused by congestion in the networking, and bit errors in random or burst fashion due to fading, interference etc. in the wireless communication. Therefore, additionally controlled redundancy in the form of channel coding as well as comprised in the source coding is required to combat the hostile channel environments. The objective of the error resilient image coding and transmission is to balance the source coding efficiency and the error combating ability in order to achieve the best end-to-end performance.; In this dissertation, we start by reviewing the principles of image coding, characteristics of communication channels, and the existing schemes for joint source and channel coding design. Then, we focus our research on four categories in detail. First, we present a channel optimal non-uniform trellis coded quantization scheme, in which the quantization parameters and the bit allocation among different subbands are designed based on the given channel condition and channel coding scheme. Secondly, we introduce the concept of multiple bitstreams robust image coding and transmission with three variable-length coding-based schemes. Limited error propagation is achieved. Unequal error protection is designed according to the bit contribution to PSNR and/or visual quality. Thirdly, we propose a completely novel product coding and recurrent alternative decoding scheme that includes several innovative components such as coding structure, constrained Viterbi algorithm, and recurrent alternative decoding etc. Fourthly, we develop a trellis-based technique for blindly joint channel estimation and signal detection using list and conventional adaptive Viterbi algorithms. Promising, and some currently best results of these proposed schemes have been achieved. Finally, we conclude this research and give a perspective of the future work. The innovative schemes in this dissertation can be extended and lead to several new research directions. |
