A Real-Time Video Codec For Telecommunication At Low Birtate

• Main Author: 瞿青文
• Other Authors: Huang, Shing-Tsaan
• Format: Others
• Language: en_US
• Published: 1996
• Online Access: http://ndltd.ncl.edu.tw/handle/44263209716240685530

碩士 === 國立清華大學 === 資訊科學學系 === 84 === 　　In this thesis, a real-time video codec for telecommunication at low bitrate (< 64 K bps) is presented and developed by software only. Since the compression time and the coded video quality are tradeoff at a fixed compression rate, the goal of our codec is to achieve the real time criterion as well as maintain a tolerable, stable video quality at a target bitrate and frame rate. Our codec design is based on the H. 261 standard, and guarantees compatibility with other H.264-type codecs. Although another standard, H.263, can provide higher quality than H.261 at the same compression ratio, but it needs much more computational costs, thus the real-time constraint of the H.263 encoder is hard to achieve by software only. However, some less-complexity techniques of H.263 are included and modified in our codec to improver the performance that also generates a regular H.261 bitsteam. 　　The tricks of the trade to reduce the computational complexity of our codec are discussed in the thesis, which furthermore can be applied in designing a H.263-type codec in the future. First, the MTYPE (Macroblock TYPE) classifier and the ME (Motion Estimation) are linked together to skip redundant searches of the ME. To further reduce the computational cost of the ME, a new ME scheme is proposed, which guarantees robust speed at the same level of video quality. A simple but fast buffer (rate) control associated with an efficient quantization is used. The simulation results show that it regulates well at a given target bitrate. Finally, two optional enhanced modes, SCF (Skipping / Compensating Frame) and QO (Quarter Operation), are introduced in our codec to achieve the real time criterion on some low-speed processor, such as 486 or lower.Both two options can be used together or separately. Simulation results indicate that our encoder with the SCF and QO options together saves about40% time, but losses about 0-3 dB in PSNR, depending on the video sequences.