Optimization And Implement Of HEVC High Definition Real-time Encoder Based On Multi-core Platform

With the development of multimedia technology and people's increasing demands on video image quality,high-definition and ultra-high definition video applications become increasingly popular.How to enlarge the existing low-resolution video to high-resolution video image and to maintain image quality,that is,super-resolution zoom technology,become research hotspots in recent years.This paper focuses on two aspects of super-resolution algorithm based on local self-similarity model,including improving the image quality after super-resolution and accelerating the algorithm performance with the help of GPU realization.Firstly,the existing local self-similarity based superresolution algorithms result in "Painting Effect" issues,so we propose a subpixel interpolation and a block matching method to solve this problem.Then the enlarged images look more natural especially when focusing on the image texture.Secondly,the present super-resolution algorithm tends to enhance the noise while image scaling,thus this paper proposes a scheme which utilizes the high-frequency block's matching information to determine how to add the high-frequency block.This will not cost extra excutive time but make the image clearer and cleaner.Since the super-resolution algorithms based on local self-similarity model use iterative enlargement and get the high resolution image pixel by pixel,the computation amount is huge.Especially when carring out 4K(Ultra HD)zoom,the processing speed is difficult to meet the pracatical application.To solve this problem,this paper take advantage of GPU hardware features and realize the GPU's parallel algorithm design to achieve more than one hundred times acceleration.