Object-based video coding with dynamic quality control

摘要

This paper presents a novel MPEG-4 priority-based multiple video objects coding system integrating User Visual Attention Model (UVAM) and Dynamic Bit Allocation (DBA) rate control mechanism. In the proposed system, the importance of individual objects is defined automatically and the coding parameters are adaptively adjusted so that the perceptual quality of important objects is maximized. The first contribution of the paper is an object-based UVAM utilizing a two-level structure and a blob extraction algorithm. Following that, two DBA algorithms are designed for object-based coding quality control in which weighted bits are allocated to prioritized single or group of video objects. Experimental results confirm that the defined automatic object-based priority is consistent with the observation of human visual system and the coding quality is maximized for the most important object and group of objects. In addition, the proposed system provides object-based coding flexibility and improved efficiency of utilizing coding resources. It offers key techniques to ensure the quality of services for real-time content-based video coding and streaming systems.