Error concealment for robust image and video transmissions on the Internet

Coding and transmission of images and videos have been a popular but very challenging topic. Traditional coding algorithms are usually designed to optimize compression ratio in an error-free environment but not for the current Internet that is only a best-effort, packet-switched and unreliable network. This conflict presents a number of challenges for high-quality image and video transmissions.; Information loss and bandwidth limitation are two major factors that affect the quality of video streaming. In this thesis, we have developed various error concealment and reconstruction-based rate control schemes to address these two issues. First, we have proposed a sender-receiver based approach for designing a multiple-description video coder that facilitates recovery of packet losses. Second, we have studied artificial neural network-based reconstruction algorithms for compensating nonlinear compression losses due to multiple-description coding. Third, we have employed syntax-based packetization and decoder-side feedback for reducing propagation losses. Last, we have incorporated the reconstruction process in the design and evaluation of rate control schemes.; Likewise, delay and packet loss are two primary concerns for image transmissions. Existing approaches for delivering single-description coded images by TCP give superior quality but very long delays when the network is unreliable. To reduce delays, we have proposed to use UDP to deliver sender-receiver based multiple-description coded images. To further improve the trade-offs between delay and quality of either TCP delivery or UDP delivery, we have investigated a combined TCP/UDP delivery that can give shorter delay than pure TCP delivery and better quality than pure UDP delivery.