Presenter

Nikolaos Deligiannis - ETRO-VUB

Abstract

In the modern wireless age, lightweight multimedia technology stimulates attractive commercial applications on a grand scale as well as highly specialized niche markets. In this regard, the design of efficient video compression systems meeting such key requirements as very low encoding complexity, transmission error robustness and scalability, is no straightforward task. The answer can be found in fundamental information theoretic results, according to which efficient compression can be achieved by leveraging knowledge of the source statistics at the decoder only, giving rise to distributed, or alias Wyner-Ziv, video coding. This dissertation engineers efficient lightweight Wyner-Ziv video coding schemes emphasizing on several design aspects and applications.

The first contribution of this dissertation focuses on the design of effective side information generation techniques so as to boost the compression capabilities of Wyner-Ziv video coding systems. To this end, overlapped block motion estimation and probabilistic compensation, a novel technique that performs advanced multi-hypothesis motion-compensated prediction at the decoder, is proposed. Using auxiliary (i.e., hash) information sent to the decoder, the proposed technique triggers the design of a novel efficient codec featuring very low encoding complexity and operating without a feedback channel. Adding a transform domain Wyner-Ziv coding layer with a feedback channel yields a novel codec that outperforms state-of-the-art Wyner-Ziv codecs. What is more, tailored to a novel hash design, an innovative modified version of the proposed technique is included in a new efficient hash-based Wyner-Ziv architecture. Furthermore, when coupled with an alternative side information creation method, the proposed technique enables side information refinement after decoding critical information, thereby further improving the performance of traditional Wyner-Ziv video codecs.

The second contribution of this dissertation constitutes the introduction of a novel correlation channel modeling concept that expresses the correlation noise as being statistically dependent on the side information. Compared to classical side-information-independent noise modeling adopted in traditional Wyner-Ziv coding solutions, it is theoretically proven that side-information-dependent modeling improves the compression performance. Anchored in this finding, a novel algorithm for online estimation of the side-information-dependent correlation channel parameters is presented. The proposed algorithm enables bit-plane-by-bit-plane successive refinement of the channel estimation leading to progressively improved accuracy. Experimental results corroborate the theoretical coding gains brought by the side-information-dependent model and demonstrate the superior accuracy of the proposed online channel estimation algorithm over state-of-the-art approaches.

The third contribution of this dissertation intends to bridge the gap between distributed video coding and its practical applications. A keynote contribution in this direction is the expansion of the application domain of Wyner-Ziv coding from conventional video to medical imaging. Wireless capsule endoscopy in particular, which is essentially wireless video capturing and transmission by a pill, is proven to be a promising application field. Driven by such applications, a new Wyner-Ziv system, which generates a hash as a downscaled and subsequently intra coded version of the encoded frame, is proposed. Based on this hash, side information generation can adapt even to extreme spatial variations in temporal correlation, often appearing in endoscopic video content. By enabling low encoding complexity and scalable coding and by delivering improved compression performance compared to the state-of-the-art, the developed codec constitutes a strong candidate for lightweight (medical) imaging applications.