Video transcoder architectures for networked multimedia applications

For the deployment of multimedia services over heterogeneous networks, it is very desirable that a video delivery system be able to dynamically adapt the bit-rate of a video bit-stream to the time-varying network conditions, such as the available bandwidth. A video transcoder which can adapt the bit-rate of a video bit-stream and support various content. manipulations, error-resilience, and video watermarking insertion will improve the overall performance of multimedia systems.; In this dissertation, we investigate various issues regarding transcoder architectures for networked multimedia applications. We first investigate existing fast architectures and analyze potential problems and limitations. We show that due to inherent but often ignored assumptions in the derivation, the existing fast architectures have severe limitations in practical applications. We also propose a fast transcoder scheme which does not have the limitations and problems of the existing fast transcoder architectures and can achieve drift-free performance with similar processing speed as the existing fast transcoders. In the proposed architecture, drift-error is prevented by adopting a pixel-domain cascaded structure. To reduce the computational cost in transcoding, several stages of computational blocks, such as Discrete Cosine Transform (DCT), Inverse DCT (IDCT), quantization, and inverse quantization are optimized.; We also extend the transcoding function to applications. of a logo insertion. We consider the case when transcoding is necessary in the logo insertion process to reduce the bit-rate.; As many multimedia services are deployed in the Internet environment, the video delivery system requires a more intelligent and sophisticated bit-rate adaptation scheme with a high transcoding ratio because of the wide range of operating bandwidths of the current Internet infrastructure. Down-conversions of frame-rates or picture resolutions can be effectively used for achieving a high transcoding ratio while maintaining an acceptable video quality. For this temporal and resolution transcoding, we propose a fast-motion vector refinement scheme to obtain the outgoing motion vectors from the motion vectors extracted from the incoming bit-stream. Simulation results show that significant computational saving is possible while achieving almost the same video quality compared with that of using full-scale motion estimation.