Low Complexity H.264 Intra MB Coding

The H.264 video coding standard is highly efficient and at the same time highly complex. The complexity of encoding is substantially higher than prior standards such as H.263 and makes H.264 video encoding on mobile devices expensive. The few existing encoding solutions on mobile devices restrict the features used there by sacrificing quality for complexity. Fast encoding algorithms are necessary to fully exploit the advanced compression features offered by H.264. In this paper we present a novel machine learning based approach to reducing the complexity of intra MB coding. The results show that machine learning has a great potential and can reduce the complexity substantially with negligible impact on quality. The results show that the proposed method reduces encoding time to about 1/3 of the reference implementation with a negligible loss in quality.