Efficient Low-Complexity MPEG-2 to H.264/AVC Video Transcoding System Architectures and Optimization Techniques Susceptible to Dynamic Partial Reconfiguration
Abstract
The variety of available coding standards introduces the need for efficient transcoding techniques. Transcoding between MPEG-2 and H.264/AVC is of particular interest due to the wide use of MPEG-2 in TV broadcasting and H.264/AVC in mobile multimedia applications. In this thesis the effort has been focused on developing and implementing transcoding systems characterized by low complexity and high throughput. Novel lowcomplexity architectures of MPEG-2 to H.264/AVC intra transcoding system are proposed. These architectures introduce complexity reductions such as wavefront scanning order and transform mode decision to perform real-time transcoding for a variety of resolutions. The computationally demanding yet regular units of the MPEG-2 decoder and the H.264/AVC encoder stages have been implemented with a high level of parallelism. A wavefront macroblock level scanning order based on on-the-fly processing of a number of consecutive macroblocks exploits the block independence in intra prediction. On the other hand, the architecture with mode decision algorithm for choosing between two intra prediction algorithms removes the dependency between intra prediction algorithms. Achieved results in terms of performance represent a significant reduction of minimal required frequency compared to the state of the art for resolutions CIF, SD and HD1080p. Furthermore, the transcoder implementation with wavefront processing supports throughput that corresponds to 4320p at 30 fps, whereas the second implementation can process video sequences up to 2160p at 30 fps. An aspect of the proposed transcoding systems being explored is a flexibility feature controlled by the user. The idea of adaptive transcoding system relies on the use of the partial reconfiguration. The partial reconfiguration, as an optimization technique, provides increased functionality of the system on the limited hardware resources. It is achieved by swapping a number of video processing algorithms or modifications of the same algorithms in time. As an intermediate step in implementing a flexible transcoding system by the use of dynamic partial reconfiguration, video processing systems implementations such as adaptive video filtering system, adaptive H.264/AVC quantization system and adaptive DCT transform system for multi-standard processing are built within the heterogeneous system. The adaptive video filtering system implementation uses the dynamic partial reconfiguration for swapping between filter functions such as edge detection, de-noising and thresholding functions. The process itself shows advantages such as resource saving but also limitations such as high reconfiguration time and long development process.