Improved rate control via rate-quantization modeling with kalman filter

Abstract

In this paper, we propose a rate control framework for H.263+ using an adaptive rate-quantization (R-Q) model. A characteristic rate function parameterized by a specified quantization parameter can be approximately a linear function of macroblock (MB) activity. During encoding, Kalman filter simultaneously refines the slopes of R-Q characteristic line to trace the change of R-Q relation between two successive clustered MBs (COBS), and the quantization parameters of MBs located in the same COB can be determined based on the latest refined R-Q model. For satisfying real-time demands, the proposed scheme employs a fast method of progressive MB mergence to determine the range of a COB. In the experiments, our framework can obtain higher objective and perceptual qualities over TMNB module. Keyword: H.263+, Rate-Quantization Model, Kalman filter.