Encoding Optimizations Based On Motion Analysis For High Efficiency Video Coding(HEVC)

With the development of multimedia technology,upgrading of digital devices and increase of network bandwidth,there are increasing requirements for the high definition(HD)digital image and video services.Large number of HD videos are also spreading through the Internet in our daily life,e.g.,TV shows or movies for entertainment,video conferencing for working,and remote education for study.First,it is noted that the resolution of video is increased from 176×144 at the early ages to currently commonly used HD of 1920×1080(1080p),and even to the ultra HD of 8K×4K in the near future.Then,the content category is becoming much richer.Besides the traditional camera-captured content,screen content is becoming increasingly popular in recently years which may contain texts,graphics,animation and camera-captured content in the same picture.To reduce the cost of video storage and transmission,efforts are required to improve the coding efficiency.For the practical use,a high coding speed is also preferred.Both the requirements need proper optimizations to video encoder.High Efficiency Video Coding(HEVC)is the state-of-the-art video coding standard,which doubles the coding efficiency of its predecessor H.264.HEVC extensions have also developed special coding tools for the screen content coding and further improve its coding efficiency.Based on the analysis of motions in video content,this dissertation focuses on the algorithm improvement and implementing optimizations for both the coding efficiency and speed of HEVC encoder.Motion estimation is one of the most important coding tools in traditional video encoding as well as the state-of-the-art HEVC encoding.It helps to find the block with identical or similar content to make a prediction by the motion search among neighboring pictures.The performance of motion estimation module will have a great impact on both the coding efficiency and speed,whether it is for the video encoding of camera-captured content or of screen content.More importantly,motion estimation not only helps to find a good prediction,but also gives insights into the changing characteristics of video contents by the motion analysis,which can be used to guide the encoder optimization on both the fast encoding and coding efficiency improvement.The main innovations and contributions of this dissertation are as follows.(1)For the HEVC coding speed optimization,this dissertation proposes a fast HEVC encoding scheme with the cooperation of Graphics Processing Unit(GPU)and Central Processing Unit(CPU).There are two main innovations.First,a parallel motion estimation scheme on GPU is proposed.The proposed scheme not only improves the coding speed with the parallel processing on GPU but also can compensate the coding loss from the absence of motion vector prediction(MVP)in the parallel processing.Then,with the GPU returned motion information of all blocks,this dissertation further proposes a fast encoding optimization scheme on CPU,which greatly decreases the encoding complexity.Combined with the Wavefront Parallel Processing(WPP)on CPU,the proposed fast encoding can achieve 30~40 times of speedup on 1080 p videos with only 3.1% coding efficiency loss with the cooperation of GPU and 16 cores CPU.(2)For the coding speed optimization of HEVC screen content coding(HEVC-SCC),a fast hash-based motion estimation method with the hierarchical hash calculation is proposed.There are three main innovations.This dissertation has performed a detailed analysis on the existing hash-based motion estimation and finds that the high complexity of hash table generation module in the HEVC-SCC reference software is due to the great computation redundancy in the current hash calculation method.To overcome the high complexity problem of the existing hash table generation method,a hierarchical hash value calculation and a hierarchical hash block availability checking are proposed so that the hash table generation as well as the hash-based motion estimation can be performed with much lower complexity,which are the first two innovations.Combining the first two innovations,the proposed method decreases about 77% hash table generation time and results in 12% and 16% encoding time reductions for random access(RA)and low-delay B(LB)coding structures,respectively while the coding efficiency remains unchanged.Because of the effectiveness of the proposed method,it has been adopted by the HEVC-SCC standard and integrated to the latest reference software of HEVC-SCC.Besides the first two innovations,this dissertation also proposes a parallel implementation of the proposed hash table generation on GPU.With the proposed hash-based motion estimation,x265 encoder can compress screen content videos with average 11.8% and 14.0% bits saving(coding efficiency gains)for RA and LP coding structures,respectively when hash tables are generated in parallel on GPU.Furthermore,the real-time encoding for 1080 p videos can be achieved.(3)For the coding efficiency optimization of HEVC-SCC,this dissertation proposes a weighted rate-distortion optimization scheme.There are two main innovations.First,this dissertation analyzes the three typical cases of content motions in screen content and observes that different blocks in one picture will have different impacts on the encoding of the following pictures due to the varying motions.To take the impacts of different blocks of current picture into the encoding optimization,this dissertation proposes a block distortion weight estimation method based on motion estimation.Then,with the resulting weights in spatial domain,a detailed implementation of the weighted rate-distortion optimization in both the spatial and transform domain is proposed to improve the coding efficiency.Compared with the HEVC-SCC reference software,10.1%,14.5% and 2.2% on average and up to 25.7%,39.8% and 4.6% bits saving can be achieved by the proposed scheme for RA,LB and all intra(AI)coding structures,respectively.All the works above are based on the motion estimation and analysis of video content characteristics,HEVC encoding optimizations are then performed accordingly.The experimental results verify that the proposed methods based on motion analysis are effective in HEVC encoding optimization for both coding speed and coding efficiency improvement.