The Optimization Of Several Key Technologies Of High Eficiency Video Coding

Video Coding Experts Group(VCEG)and Moving Picture Experts Group(MPEG)jointly developed the next generation standard named High Efficiency Video Coding(HEVC)in February,2013 to meet the increasing demands on the quality of video,especially the growing popularity of high/ultra-high resolution video applications.The main goal of the HEVC standardization is to enable significantly improved compression performance relative to existing standards——in the range of 50%bit-rate reduction for equal perceptual video quality.However,excellent coding performance is at the cost of computational complexity,which is a huge challenge for the application and development of HEVC.Therefore,it is very important to study how to reduce the computational complexity at a negligible loss of video quality.We first review the development of video coding standards and make a study of research progress on HEVC at home and abroad in recent years.Then after introducing the framework of encoding,we talk about the key technologies in detail,such as the flexible block partition structure,intra prediciton,inter prediction and so on,which is useful for the following optimization steps.We perform the improvement in two aspects:We have made a study of the parallelism base on homogeneous multi-core architecture.Then we propose a parallel strategy based on WPP after data dependency analysis for HEVC encoder.We use the concept of critical path to demonstrate the correctness of the optimization of the parallel strategy and then figure out the ideal parallelism and average parallelism based on this theory.Finally,we implement the WPP parallel algorithm on official reference software HM with the help of POSIX thread.Experiments show that the algorithm can enhance the speed of encoding without affecting the video quality.We find that there are several features presenting high correlation with the best splitting decision of CU.As a result,the splitting decision is modeled as a classification problem in data mining.We choose decision tree to solve this problem.First of all,we divide the standard test sequences into two subsets:one is training set and the other is testing set.We build a decision tree using the data in training set.The nodes except leaves on the tree are the features that present high correlation with the best splitting decision while the leaves denote the final splitting result.At last,we implement the decision tree on official software HM and test the algorithm on the testing set.Experiments show that the fast algorithm based on decision tree can efficiently enhance the encoding speed at a negligible cost of 0.3%in terms of BD-rate.At the end of this paper,we summarize the work of this paper and outlook the key points of the research in the future.