| Summary: | 博士 === 國立成功大學 === 資訊工程學系碩博士班 === 97 === Bit-rate control plays a major role in video coding and multimedia streaming. Well-designed
bit-rate control mechanisms are able to 1) efficiently adjust encoding quantizers for achieving fine visual qualities in subject to encoding bit budgets and 2) adjust the transmission rate for avoiding network congestion in subject to the predicted available bandwidth. In the dissertation, bit-rate control
mechanisms for H.264 video coding and ransmission over heterogeneous networks are designed and analyzed, including three critical issues: bit allocation and quantizers decision (at the encoding stage), bandwidth management and scalable audiovisual streaming synchronization (at the streaming stage).
For encoding-stage bit-rate control, H.264 bit-rate control using a 4-D perceptual quantization
modeling (PQrc) is proposed, including two major encoding modules: the perceptual frame-level
bit-allocation using a 1-D temporal pattern and the macroblock-level quantizer decision using a 3-D rate pattern. The temporal pattern is used to predict frame complexity and determine proper bit budgets further. The rate pattern is depicted as a bit-complexity-quantization (B.C.Q.) model, in which a tangent slope of a B.C.Q. curve is a piece of unique information to find a proper quantizer.
For newly generated video clips, the B.C.Q. model can be updated continuously using a weighted least square estimation.
For streaming-stage bit-rate control, in order to adapt the transmission rate of streamed ondemand videos to the time-varying channel bandwidth, predictive video-on-demand (VoD) bandwidth management and a feedback-based buffer control scheme are required. According to the measured information of packet round-trip-time, loss-rate, delay jitter and received bit-rate, an improved Kalman filter is adopted to predict an available channel bandwidth recursively and to determine a proper transmission rate in consideration of buffer fullness of a decoder. The optimal parameters of the Kalman filter, e.g., a transition matrix and error covariances, can be initialized, converged and adapted to
characteristics of the current network.
In addition to the aforementioned bandwidth management, scalable audiovisual streaming synchronization is also a main issue we concern. Considering characteristics of heterogeneous networks and device capabilities, streamed audiovisual content is requested to be scaled with the proper spatial/temporal resolution, format and quality levels. Past technologies of scalable audiovisual coding, e.g., Fine-granular scalability (FGS) and bit-sliced arithmetic coding (BSAC), are used to segment video and audio data into one base-layer and multiple enhancement-layer bitstreams. With the advantages
of scalable audiovisual coding, a de-jitter procedure, a conditional retransmission mechanism
and a playout synchronization mechanism are designed to transmit hybrid scalable (multi-layered) audiovisual bitstreams in consideration of the result of a network bandwidth adaptation and distinct decoding time-complexity.
Finally, experimental results show that 1) the proposed H.264 rate control can keep stable buffer fullness and improve the SNR quality and control accuracy effectively in comparison with H.264 JM10.2; 2) the proposed bandwidth management using the improved Kalman filter can obtain more precise and stable estimation results of bandwidth in comparison with pathChirp and decrease the packet loss-rate when data are streamed in real wired/WiFi/3G networks; 3) the proposed scalable
audiovisual streaming synchronization can eliminate effects of the delay jitter, increase the in-time decoding ratio and perceive audiovisual playout smoothly.
|
|---|
