Power and distortion optimized video coding for pervasive computing applications

This dissertation investigates video encoding schemes for pervasive computing applications that must ensure low power consumption in addition to high compression efficiency. The contribution of the dissertation is the formulation of a theoretical problem that captures the joint optimization of power and distortion in video coding. The study of the complexity distribution of typical video encoders helps to develop a complexity-scalable video encoding architecture that includes several control parameters to adjust the power consumption of the major modules of the encoder. An analytic framework to model, control and optimize the power-rate-distortion is developed, which facilitates the development of optimization schemes to determine the best configuration of the complexity control parameters according to either or both the power supply level of the device and the video presentation quality. The dissertation proposes complexity control schemes that dynamically adjust the control parameters. Using extensive simulations on an instruction set simulator, the accuracy of the model, and quality of the optimization schemes are investigated. For additional performance improvement, we propose algorithms that exploit the video content to reduce the power consumption and improve the video quality. This is done by obtaining and maintaining the "motion history" of a video sequence in a hierarchical fashion. By adaptively adjusting the complexity parameters according to the motion history information gained from the video sequence, the power is saved when the scene has little motion and consumed when the motion activity increases. Extensive experiments have been performed to show the validity and merits of the proposed techniques.