Optimizing Motion Estimation for H.264 Encoding

• Mateus Krepsky Ludwich UFSC
• Antônio Augusto Fröhlich UFS

Resumo

Around 90% of the total encoding time of raw video into H.264 is spent in block-matching, a stage of Motion Estimation. In this paper we combine two block-matching optimizations that yield significant performance improvements on Motion Estimation: 4:1 subsampling by macroblock and truncation of the two less significant bits per sample. When applied to the JM Reference Encoder, our strategy showed an average speedup of 2.64 times in total encoding time with a small loss of quality (less than 0.5 dB).