Quantizer Design to Exploit Common Information in Layered and Scalable Coding

This paper considers a layered coding framework with a relaxed hierarchical structure, tailored to serve content at multiple quality levels, where a key challenge is the conflict between coding optimality at each layer and efficient use of storage and networking resources. The prevalent approach of storing and transmitting independent copies for each quality level, is highly wasteful in resources. The alternative of conventional scalable coding incurs the notorious “scalability penalty” at the enhancement layers, due to its rigid structure. The approaches pursued in this work involve a layered coding framework, wherein information common to one or more subsets of the quality levels is first extracted and transmitted, and then complemented by individual (quality level specific) bit streams. This framework ensures that no redundant or irrelevant information is sent to any decoder, enables achieving all intermediate operating points between the two extremes of conventional scalable coding versus independent coding, and hence mitigates the layered coding penalty. Joint design of common and individual layers ensures that all extracted common information is fully usable by the target decoders, as needed to approach optimality. Simulation results for practically important sources, confirm the superiority of the proposed framework.