Source and channel coding of images for noisy channels

An important problem is how to efficiently encode images for transmission over limited-bandwidth channels while maintaining good decoded image quality and resilience to channel impairments. This thesis investigates the problem of robust image transmission over memoryless, fading, and packet erasure channels.; First, a system consisting of a high-performance embedded wavelet coder followed by a concatenated channel code is proposed for transmission over memoryless channels. The system maintains the progressive nature of the source coder, has reasonable complexity, and exceeds the performance of previously known systems. Error resilience and progressivity are further explored through the problem of channel code parameter (block length and channel code rate) optimization for progressive transmission. A general performance measure for evaluating progressive image coding quality is proposed and a dynamic programming optimization algorithm is presented.; For fading channels, the concatenated code proposed for memoryless channels is extended to a product code structure. The new code shows improved performance on fading channels and slightly improved performance on memoryless channels. It also has variable amounts of delay, depending on the level of the channel noise, allowing the progressive performance under good channel conditions to remain high. Modifications to the source coder to increase its inherent error resilience in combination with explicit channel coding are considered for varying channels, including those that introduce both bit errors and packet erasures. A general performance measure is proposed to facilitate the optimization of system parameters. Improvements are shown in terms of robustness to a range of channel conditions.; Finally, methods for introducing an adjustable amount of source redundancy are explored in the context of erasure channels. The erasure channel can be error detection is possible, so the methods are broadly applicable. A multistage coding structure is proposed which offers much flexibility, even in terms of the underlying compression algorithms, as well as low complexity. Extensions of multiple description scalar quantizers in combination with erasure correcting codes are proposed as another method.