Very low bit rate video coding

Recently a novel pattern-based video coding (PVC) technique was proposed to exploit a special type of intra-block redundancy by focusing on moving regions using pattern templates to achieve superior coding efficiency over the H.263 standard. Although the variable block size (VBS) mode of the latest H.264 standard partially exploits the same kind of redundancy, the PVC has been proved superior in terms of both computational complexity and image quality, especially at very low bit rate video coding. However, the existing work on PVC could reduce the bit rate for the same image quality compared to H.264 using only the VBS mode by not more than 5%. This thesis contributes to the acceptance of the PVC concept as a potential additional mode for future H.26X video coding standard by improving relevant rate-distortion performance, adaptability, computational complexity, scalability, optimality, and transcoding issues. Different aspects of the preliminary ideas are enhanced by modifying moving region definition, pattern code book size, pattern matching criteria, and pattern identification codes. A real time pattern selection algorithm is developed by intuitively engaging a low complexity pattern matching criterion along with the existing high complexity one. This algorithm ensures the maximum possible image quality for a given pattern code book with the least computational complexity. The concept of PVC is then extended to the content-based coding paradigm by developing an efficient heuristic to generate arbitrary-shaped patterns of different sizes from the video content. For the sake of completeness, this thesis also presents a greedy heuristic for reducing the size of a large pattern code book to a manageable size so that pattern identification overhead is kept limited. Experimental results show that the real time pattern selection algorithm reduces the bit rate as much as 20% with only about 30% computational time for real-time streaming; while the arbitrary-shaped PVC with three pattern sizes reduces the bit rate as much as 35% for stored video, for the same image quality compared to the H.264 standard using only the VBS mode.