Research On An Comprehensive Algorithm Based On Texture Feature For HEVC

ITU-T video code expert group(VCEG) and ISO/IEC motion picture expert group(MPEG) came up with a new generation of video coding standard HEVC(High Efficiency Video Coding) in 2013. HEVC introduces more progressive coding technology based on H.264/MPEG-4 strandard with traditional hybrid coding framework, including providing specific units for different stages of intra coding process (introduction of CU、PU、TU), more detailed division of angle for intra angle predition modes, efficient Context Adaptive Binary Arithmatic Coding (CABAC), import of a new gernaeration of motion blending technology and so on. Introduction of these new technology, compared with H.264/AVC standard, dramatically increase efficiency of video compression with an reduction of 50% in bitrate on the premise of guaranteeing pictures’ quality.Meanwhile, computational complexity of HEVC significantly grows in coding process with a series of new technology, it seems that real-time performance of encoding can not be guaranteed. This thesis starts with two aspects as follow to cut down on encoding time. At first in intra prediction, this thesis aims at Most Probable Mode(MPM) mechanism which is newly introduced in HEVC intra mode selection process, classifies frames according to texture feature, then simplifies every types of frames’ mode selection process in specific ways, respectively, to make MPM more detailed, more adaptable, reducing encoding time at the same time. Secondly, in inter prediction, an algorithm which stores depth information of each frame is proposed in this thesis. This algorithm utilizes these stored depth information to simplify depth traversing process in the prediction process of subsequent frames to attain the goal of reducing encoding time. Meanwhile, to guarantee the precision of prediction, Reference Frame Selection(RFS) mechanism and influence of Quantization Parameter are also taken into account. So the algorithm proposed in this thesis is an comprehensive optimization based on series of video coding machanism.Algorithms proposed in this thesis are implemented in code in HEVC standard reference software HM version 12.0. As is seen in experimental results, the intra optimization algorithm proposed in this thesis cut down about 35% on encoding time in average, limiting the growth of bitrate within 1.5% at the same time. And the comprehensive depth dividing algorithm proposed in this thesis limits increase of bitrate in 2% and causes over 30% reduction on encoding time. When the type of frame being encoded is B-Slice or P-Slice, two proposed algorithms can be applied together in coding process according to relevant settings of encoder configuration file. Two proposed algorithms present strong adaptability for high resolution video coding with the fact that bitrate does not increase substantially as video resolution is getting higher and higher, which ndicates a bright future of these algorithms based on texture features for high definition video.