A novel fast mode decision algorithm for H.264/AVC based on probabilistic learning

This paper presents a fast mode decision algorithm for H.264/AVC encoding optimization. The algorithm is based on probabilistic learning. Spatial correlation between neighboring macroblocks is exploited by K-means clustering to predict the best candidate mode for the current macroblock. Early termination strategy is designed for skipping unnecessary mode checking. Extensive experiments shows that the proposed algorithm can achieve 50%-80% encoding time reduction compared to JSVM9.18, and only yield a negligible PSNR loss and little increment in total bits and bit rates.