Error-resilient Image And Video Coding Over Lossy Networks

Abstract

The rapid growth of the Internet and increasing bandwidth in the past decade has heightened the interest in robust transmission of image and video signals over error-prone links. Recent literatures highlight the multiple description (MD) coding as a promising approach to solve this problem. MD coding is a source coding technique that provides a graceful quality degradation in the presence of packet losses during transmission. Topics of interest to construct MD coders with high coding efficiency and robustness are studied in this thesis. First, to minimize the overall distortion of reconstructed image, an edge-adaptive estimation algorithm is employed, which can capture the arbitrarily-oriented edges around the missing samples. Remarkable gains are obtained compared to the conventional method in the literatures. Second, a feature-oriented MD coding scheme is proposed by adopting this estimation algorithm to smartly identify the coefficients that are sensitive to packet loss. The redundancy is then inserted to each channel by jointly optimizing the subsets of sensitive coefficients and the related quantizers. Significant improvements are achieved on both coding efficiency and transmission robustness. Finally, an optimal redundancy allocation algorithm is developed for the drift-free MD video coding system, which can efficiently adjust the source coding rates for the drifts and the introduced correlations in a coordinated fashion.