Embedded video compression for high dynamic range video

Abstract

Modern digital TV-systems encompass digital demodulation and required audiovisual signal processing incorporating an uni ed-memory architecture. The bottleneck of such a System-on-Chip (SoC) is the connection to the external memory. As the video signal consumes the majority of the involved bandwidth, embedded video compression allows bandwidth reduction, enabling cost e ective systems. The embedded video compression must be visual lossless and cost e ective. This requires a high e cient de-correlation step to limit the amount of quantization, thus avoiding a decrease of picture quality. In this thesis, video characterization has been conducted, resulting in novel video de-correlation concepts. Based on these concepts, a DCT-based codec exploiting symmetrical features, structure repetitions and prediction techniques in the DCT transform domain has been developed. From the results it can be concluded, that symmetrical features turn out to be an e cient method to compress video, preserving the demanded high picture quality at acceptable computational cost.