Stable Buffer Control and Effectively Bit-Allocation with Kalman Filtering Algorithm

• Main Authors: Chia-Hui Chao, 趙家慧
• Other Authors: Din-Yuen Chan
• Format: Others
• Language: en_US
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/79720211951929075122

碩士 === 義守大學 === 資訊工程學系 === 88 === Bit-rate controlling plays a significant role for stable real-time visual communication. The paper introduces an efficient rate-buffer control algorithm for the H.263 standard to achieve well spatial-temporal quality control according to the video scene content. In the proposed scheme, a Kalman filtering predictor is a major part to estimate the encoded bit, assign target bit-rate per frame and use prediction error feedback to stabilize buffer content. To handle well tradeoff of temporal and spatial qualities, we use long-term and short-term frame skips mechanisms where the long-term one makes decision on the characteristics of windowed pictures. And, to use the incorporation of encoded-bit predicting via Kalman filtering and the short-term frame skips decision, not only the proposed rate buffer controller can precisely avoid overflowing the transmission buffer, but also obtain proper bit-budget for each picture. After selecting an appropriate rate-distortion model, of which parameters are the functions of bandwidth. Using off-line regression analysis and the statistical average according to the occurrence probabilities of quantization parameters can primarily yield the model parameters. And the suitable slice quantization parameters are determined by the energy ratio of each other slices in this picture. The macroblock quantization parameters are adjusted in a small dynamic range according to macroblocks’ activities. Further, the proposed rate buffer controller has load-balanced scheme, where more complex computation exists higher H.263 layers, such that it can be suitable for performing in the processor with the ability of parallel processes. Consequently, such a proposed rate buffer control with omni-directional consideration and reasonable complexity can be high regularly and effectively realized in practical real-time system.