Leaky Prediction Based Region-of-interest Scalable Video Coding

We propose a region-of-interest (ROI) scalable video-coding scheme based on leakyprediction (LP) for robustly transporting video over error-prone networks. It deploys a leakyfactor to weigh the two predictions, one from the constrained motion estimation within theROI layer of the reference frame, and the other from the unrestricted motion estimationin the overall reference frame. We analyze the proposed coding scheme, and prove it canreduce the error propagation caused by the background layer destruction due to erroneoustransmission, while it can guarantee the ROI layer quality throughout the video sequence.Simulation results show that the LP-based ROI scalable coding achieves up to 1dB in thereconstructed ROI layer and improves the overall frame quality as well. It confirms the errorrobustness of the proposed scheme, yet still preserves the coding efficiency. Furthermore,we investigate several parameters that are closely related to the performance of the proposedscheme.To better analyze the LP-based ROI scalable coding theoretically, we propose a rate-distortion model of the coding method, and try to formulate the correlation between leakyfactorαand bit rate R, as well as system distortion D. The RD analysis based on the proposedmodel can be used as a guideline to select the optimal leaky factor and to do rate control.Finally, a computational model for scalable visual sensitivity profile (SVSP) is proposedto extract ROI before LP-based ROI scalable coding. SVSP is a hierarchy of saliency mapsthat simulate the bottom-up and top-down attention of the human visual system (HVS). Theproposed SVSP can be directly used to augment the visual quality of spatial scalability.Furthermore, ROI-based SVC also benefits from the SVSP since perceptually prioritizedROI can be extracted by simply segmenting the SVSP via thresholds. Experiments put theROI defined by SVSP to use in the proposed leaky prediction based ROI coding.