Research On Intra Coding Algorithm Based On Optimal Depth Prediction For Virtual Reality Video

At present,the development of virtual reality(VR)technology is more and more rapid,and its application is more and more extensive.Virtual reality video is usually projected first,and then encoded with traditional video coding framework(such as HEVC).Due to the high frame rate and high resolution of VR video,traversing every size coding unit(CU)in intra-frame coding process generates more computational complexity,and the coding speed is slower than that of traditional video.When encoding virtual reality video,there is no optimization for different projection formats.This paper analyses the characteristics of VR video,optimizes the VR video coding algorithm,focuses on the fast partitioning algorithm of intra-frame coding CU,designs a fast partitioning algorithm of CU based on optimal depth prediction,and an early termination algorithm of CU partitioning based on latitude-weighted coding bits.The main work and innovations of this paper are as follows:(1)By predicting the maximum coding unit(LCU)depth to skip the calculation of part size CU,a fast partition algorithm of CU based on LCU depth prediction is proposed in this paper.The algorithm uses Sobel operator to get LCU texture features,analyses the relationship between LCU texture features and LCU depth,and uses KNN classifier to predict LCU depth.In order to make the KNN training set data more suitable for the video to be coded,the video is divided into prediction frame and training frame.The training frame is coded by the original algoritthm and the texture features and depth of the training frame LCU are recorded in the KNN classifier.The prediction frame uses the KNN classifier to predict the LCU depth,skipping the calculation of part size CU.The experimental results show that the algorithm shortens the encoding time by 37.9%and increases BD-rate by only 1.3 1%.(2)Intra-frame coding speed is improved by early termination of CU downward partitioning.According to the characteristics of virtual reality video in ERP projection format,different reg:ions of CU are weighted.A latitude-weighted algorithm for early termination of CU partitioning is proposed in this paper.The algorithm judges whether the CU needs to be divided downward according to the number of coding bits of the CU,and sets a threshold to judge whether the CU with different sizes and quantization steps needs to be divided downward.In view of the fact that the high latitude area of ERP projection format has little influence on video quality,the U is weighted according to latitude,which makes the CU of high latitude area easier to terminate partitioning.The experimental results show that the algorithm shortens the encoding time by 32.8%and increases BD-rate by only 1.30%.