| Summary: | 博士 === 國立臺北科技大學 === 資訊工程系研究所 === 105 === In current heterogeneous networks, video can be transmitted under different bandwidth requirements and delivered to various end devices with different resolutions and computing power. Consequently, single-layer coded video does not meet all of the aforementioned bandwidth and device requirements. Scalable video coding (SVC) facilitates the utilization of once-encoded video at a target bit rate, quality level, and computational capability. The main contributions of this dissertation are three folds. First, we comprehensively investigated the rate-distortion performance and computational complexity of the state-of-the-art H.264 SVC and SHVC (scalable high-efficiency video coding) standards. We proposed appropriate configuration settings for each scalability configuration to improve the performance of layer coding. Second, we investigated the bitrate assignment of H.264 SVC quality layers and the bandwidth allocation strategies of content delivery networks (CDNs) for video services. Among the three bit-rate ratios of the SVC quality layers (constant, Fibonacci, and exponential) and three bandwidth allocation strategies (stop and wait, stationary quality, and quasi-equal quality) for CDN surrogate servers, simulation results showed that the quasi-equal-quality bandwidth allocation strategy with the SVC layer sizes forming a Fibonacci sequence generally achieved the most favorable peak-signal-to-noise ratio and waiting time of end users. To further provide differentiated service for users of different service classes, we adjusted the bandwidth assignment of CDN surrogate servers by solving a linear integer programming problem that optimizes weighted average visual quality. Finally, we proposed an optimal server selection algorithm for video requests under CDNs, combing SVC (less computation but more cache storage) and transcoding (less cache storage but more computation). Simulation results showed that the proposed method effectively increased the number of services and reduced the waiting time of end users, as compared to the method that chose the nearest surrogate server.
|
|---|
