February Fourier Talks 2006

Advanced Video Coding, and the Influence of Wavelets

Lossy video coding is by now a maturing subject, albeit still not an exact science. It is largely predicated on the structure of image coding, which is a more mature subject. While no global optimization approach yet exists for the problem of image coding, in practice it has now stabilized into a 3-stage process of transform, quantization, and entropy coding. Video coding goes a step further, and incorporates a motion estimation/compensation loop, giving it both temporal and spatial decorrelation. While the follow-on stages of quantization and entropy coding have seen improvements in recent designs, it is primarily the subject of advanced spatial and temporal decorrelation that has been the focus of recent designs, especially the latest international standard: ITU/H.264 | ISO/IEC/MPEG-4 Pt. 10 (AVC), released in July, 2003, and updated July 2004. A scalable extension of the standard is due by Dec., 2006. This talk develops some of the main design features of the H.264 video coding standard, and compares it to several previous designs, and highlights some of its advanced features. Ironically, while wavelet concepts are not directly visible in the design, and DCT-based designs still prevail, the methods of lifting, first developed for wavelets, have played an important role in both the base design as well as the scalable extension. And wavelet-based video coding methods remain as competitors to these designs, ever ready to replace the DCT-based structures should they come up short. And the dream of someday achieving a powerful, fully embedded video codec lives on.