Learning Based Fast H.264/AVC to HEVC INTRA Video Transcoding for Cloud Media Computing

Abstract

Cloud video transcoding enable to convert the video standards and properties from one to another so as to adapt to different user end devices and network capacity, especially in sharing massive video contents in cloud environment. High Efficiency Video Coding (HEVC) and H.264/Advanced Video Coding are two recent high performance video coding standards that are widely used and co-existing in video industry. Video transcoding is desirable to bridge the standard gap. To effectively transcode video stream from H.264/AVC to HEVC for higher compression efficiency and meanwhile maintaining low computational complexity, a learning based fast H.264/AVC to HEVC transcoder is proposed for cloud media computing. We firstly analyze the correlation of block partition sizes between these two standards and then present a fast Coding Unit (CU) decision algorithm, in which three levels of binary classifiers are used to predict different CU sizes in HEVC intra coding and the optimal parameters are determined by statistical experiments. The experimental results show that the proposed transcoder achieves 44.3% time saving on average with only negligible quality degradation when compared with the original cascaded transcoder and is also superior than the state-of-the-art benchmarks in terms of complexity reduction and rate-distortion performance.