A convex optimization-based object-level rate control algorithm for MPEG-4 video object coding

Abstract

This paper presents a convex optimization based object-level rate control algorithm for MPEG-4 video object coding. By modifying the existing rate-distortion models, an improved empirical rate-distortion model with more flexibility is proposed to explain the experimental observations. The convexity and monotonicity of the proposed model are exploited to formulate the object-level bit allocation problem as a convex programming problem. Thus the bit allocation among the VOPs in a frame can be solved using efficient convex programming methods such as the interior-point methods, if the optimal solution exists. The object-level ROI functionality is also supported in the proposed framework by incorporating different object importance weights into the proposed bit allocation scheme. The model parameters are determined using previously encoded VOPs by means of linear regression. Simulation results show that the proposed algorithm achieves a considerably better picture quality in terms of PSNR than the MPEG-4 VM18 algorithm for the tested sequences. This suggests that it is a good alternative to those conventional approaches. The enabled object-level ROI functionality is also verified by experimental results. ©2006 IEEE.