You are here

Warning message

Attention! This event has already passed.

Redesign strategies unmasked. Insights in the architectural design process of adaptive reuse projects

Monday, 25 September, 2006 - 17:00
Campus: Brussels Humanities, Sciences & Engineering campus
Jonas Lindekens
phd defence

The ever increasing demand for efficient transmission of multimedia content over best effort networks and error-prone channels (e.g. packet networks, low-power wireless links) has fuelled intensive research in the area of robust communication techniques. In this context, Multiple Description (MD) coding is a competitive solution to overcome the channel impairments. This coding paradigm relies on generating more than one description of the source such that (1) each description independently describes the source with a certain fidelity, and (2) when more than one description is available at the decoder, one can combine them to enhance the decoded quality. This has the inherent advantage that the quality of the reconstructed data gracefully degrades with increasing probability of failure on the transmission channel. A common approach for MD coding is based on fixed-rate quantization. In this thesis we move one step beyond and investigate the novel concept of embedded multiple description quantization, wherein the source is MD coded and in the same time, each produced description is scalable. This has the advantage that MD coders employing such quantizers generate descriptions that posses all the features specific to a layered bitstream.

The first contribution of this thesis consists in a generalized method for designing uniform Embedded Multiple Description Scalar Quantizers (EMDSQ). These quantizers enable error resilience and support fine-grain rate adaptation and progressive transmission of each description. The main advantage of the proposed EMDSQ consists in their ability to provide uniform central quantizers at each quantization level. As demonstrated experimentally, this leads to state-of-the-art rate-distortion performance. In addition, a control mechanism enabling the tuning of the redundancy between the descriptions for each distinct quantization level is designed. The proposed mechanism enables the control of the tradeoff between coding efficiency and error-resilience, and provides an increased robustness by improving the error resilience in the most important layers of the embedded bitstreams. The flexibility of the proposed mechanism to control both the overall redundancy as well as the redundancy at each quantization level is practically demonstrated. Further, we extend the theoretical achievable regions of a classical MD system based on scalar quantization to the case of embedded quantization and, based on these findings and for an assumed channel model, we propose an optimal overall redundancy allocation for EMDSQ.

The second contribution of this thesis consists in the development of a new type of wavelet-based MD coding approaches. The proposed coding systems rely on high redundancy instantiations of EMDSQ and on QuadTree (QT) coding of the significance maps. MD-QT coding delivers image content over best-effort error-prone packet networks, and, due to its scalable erasure-resilient compression capabilities it is able to (1) meet the users’ requirements in terms of quality and resolution, (2) dynamically adapt the rate to the available channel capacity, and (3) provide robustness to data losses as retransmission is often impractical. Numerical simulations demonstrate that the proposed embedded MD coding systems provide state-of-the-art results for scalable image transmission over error-prone channels.