Adaptive Rate-Distortion Prediction for Multiple Reference Selection and Inter-mode Decision

Abstract

To minimize Rate-Distortion (RD) cost in video coding, RD Optimization (RDO) technique is adopted, which also brings increasingly computational complexity in Motion Estimation (ME), multiple reference selection and mode decision. In this paper, a Gaussian model based adaptive RD cost prediction method is developed to jointly speed up multiple reference selection and inter-mode decision. Besides the ability of achieving good coding efficiency with much reduced computational power, our method is also with two major advantages. First, it can achieve robust performance, including computational complexity reduction and almost intact video quality, with a wide range of video sequences resolutions and even for various scene changes; second, it could be either individually used or combined with the other efficient mode decision algorithms, and in both cases, our method could achieve remarkable performance, which is revealed by exhaustive experimental results.

This work was supported in part by the Hong Kong Research Grants Council General Research Fund, under Project 9041495 (CityU 115109) and City University of Hong Kong Grant 9610025, the National Natural Science Foundation of China under Grant 61070147/61102088, Basic Research Program of Shenzhen (JC2010052703), and the Program for Professor of Special Appointment (Eastern Scholar) at Shanghai Institutions of Higher Learning, the Program for New Century Excellent Talents in University of China under Grant NCET-10-0634, the Shanghai Pujiang Program under Grant 11PJ1409400, the National Natural Science Foundation of China under Grant 61102059, the Fundamental Research Funds for the Central Universities under Grant 0800219158.