Date of Award
Doctor of Philosophy in Electrical Engineering - (Ph.D.)
Electrical and Computer Engineering
Ali N. Akansu
Richard A. Haddad
Raashid Ahmed Malik
William Abraham Pearlman
Recent wavelet-based image compression algorithms achieve best ever performances with fully embedded bit streams. However, those embedded bit streams are very sensitive to channel noise and protections from channel coding are necessary. Typical error correcting capability of channel codes varies according to different channel conditions. Thus, separate design leads to performance degradation relative to what could be achieved through joint design. In joint source-channel coding schemes, the choice of source coding parameters may vary over time and channel conditions. In this research, we proposed a general approach for the evaluation of such joint source-channel coding scheme. Instead of using the average peak signal to noise ratio (PSNR) or distortion as the performance metric, we represent the system performance by its average error-free source coding rate, which is further shown to be an equivalent metric in the optimization problems.
The transmissions of embedded image bit streams over memory channels and binary symmetric channels (BSCs) are investigated in this dissertation. Mathematical models were obtained in closed-form by error sequence analysis (ESA). Not surprisingly, models for BSCs are just special cases for those of memory channels. It is also discovered that existing techniques for performance evaluation on memory channels are special cases of this new approach. We further extend the idea to the unequal error protection (UEP) of embedded images sources in BSCs. The optimization problems are completely defined and solved. Compared to the equal error protection (EEP) schemes, about 0.3 dB performance gain is achieved by UEP for typical BSCs. For some memory channel conditions, the performance improvements can be up to 3 dB. Transmission of embedded image bit streams in channels with feedback are also investigated based on the model for memory channels. Compared to the best possible performance achieved on feed forward transmission, feedback leads to about 1.7 dB performance improvement.
Zhao, Minyi, "Joint source channel coding for progressive image transmission" (2001). Dissertations. 492.