Abstracts Engineering

Improved standard-conforming video coding techniques

by I-Ming Pao

The goal of video encoding is to achieve the best video quality within the bit budget and computing power constraints. Due to the importance of video coding in various applications, through extensive research collaborations from industry and academia, several video-coding standards have been established. Currently, H.263 is widely used for low-bit-rate video applications (e.g. video streaming and video conferencing over the Internet). MPEG-2 is important for general high-quality applications including DVD (Digital Versatile Disc), Digital TV, and HDTV (High-Definition TV). MPEG-4 is a next generation content-based video-coding standard suitable for applications requiring video-object manipulations. To encourage continuing technology improvement, all the video-coding standards leave flexibility in the algorithms for further research. In this thesis, we present new standards-conforming video coding techniques which can offer better performance.We propose a statistical method to reduce the computation in the standard video encoders with negligible video quality degradation. This scheme can reduce computations in Discrete Cosine Transform (DCT), Inverse DCT (IDCT), quantization, and inverse quantization stages of standard video encoders. The information of the quantization parameter and the mean absolute errors of motion compensated blocks is used to perform the DCT calculation adaptively. A fast algorithm for approximating the 4 x 4 low-frequency coefficients is also proposed to further reduce the computation requirement. The simulation results show that the Peak-Signal-to-Noise-Ratio (PSNR) degradation is negligible and the required computation can be reduced significantly.To achieve better video quality, a sliding-window encoding-scheme is proposed for video streaming applications under the bit-budget constraint where the latency caused by the encoding process is not a concern. The statistical information of the video sequence in a sliding-window is collected. The information is then used to provide better bit-allocation in the encoding process. A quantized DCT coefficient selection scheme is also introduced to improve video quality in a rate-distortion sense. Simulation results show that significant video quality improvement is possible over the regular one-pass H.263 encoding.