Researchs On Fast Intra Mode Decision And Core Techniques In HEVC

With the rapid development of the technology, the people's demand on visual and auditory quality increase dramatically. As a result, HDTV ( High Definition Television Video) has attracted a lot of attention. The H.264/AVC video coding standard which delegates the second generation of video coding standard in the world can't satisfy people's demand on practical application. Both industry and academia has strong eager on the next generation of video coding. In this situation, both VCEG and MPEG which are two international organizations for standardization begin to cooperate and develop the next generation of video coding named as HEVC, and they have founded a joint organization which is called JCT-VC.Nowadays, although HEVC has not been completely finished, the whole coding structure has been basically established. HEVC still belongs to block-based hybrid video coding framework, however, it adopts more flexible coding unit to improve the coding efficiency. It includes Coding Unit (CU), Prediction Unit (PU) and Transform Unit (TU). CU is a macroblock-like unit for region splitting. The CU concept allows recursive splitting into four equally sized blocks, starting from LCU. This process gives a content-adaptive coding tree structure comprised of CU blocks. The PU is used only for the CU which is the leaf node in quadtree structure. In addition to the CU and PU definitions, there is another transform and quantization related unit, TU, whose size can't exceed that of the CU. The separation of three units allows each unit to be optimized. What's more, another distinguished difference between H.264/AVC and HEVC is that HEVC adopts larger block structure which can be as large as 64x64 samples.This paper proceed from two points: reducing complexity and enhancing coding efficiency to improve the coding effect of HEVC. First, this paper proposes a new fast intra mode decision to reduce the encoding time. The main idea is that we reduce the candidate intra modes for Rate Distortion Optimization(RDO) by exploiting the correlation between current block and its neighboring blocks. Second, before chroma intra prediction, we smooth the reference samples around the current block in order to remove the noise of the reference samples and avoid using rigid directional prediction to predict not rigid block. Lastly, we optimize the context model of the intra prediction modes in CABAC by counting the probability of 0 in each Bin.