Design And Implementation Of Parallel Algorithms For HEVC Intra Prediction Key Techniques

With the growing demand for video quality, VCEG and MPEG jointly launched high efficiency video coding (HEVC) in 2013 which goal is to save 50% bit rate compared to H.264. However, the excellent coding performance HEVC is at the cost of computational complexity dramatically increased. It’s necessary to improve coding efficiency in order to apply HEVC more widely. How to optimize HEVC coding speed has become a hot research topic.At present, two main processors in most computers are multi-core CPU and GPU. GPU is applied to general parallel computing because of its lots of computing units. Computer Unified Device Architecture --CUDA released by NVIDIA has provided a good platform for the GPU programming.In this paper, several parallel algorithms for HEVC intra prediction key techniques have been designed based on the study of HEVC standard and implemented on CUDA. First, different parallel algorithms of calculating luma and chroma predicted are designed by analyzing the data correlation between. Secondly, parallel algorithms for DCT and inverse DCT based on butterfly algorithm are designed. Thirdly, parallel algorithms for quantization and inverse quantization are designed. At last, the parallel algorithm of intra prediction is optimized. What’s more, a parallel algorithm for fast prediction of gradually narrowing the scope of modes is designed and implemented on CUDA. The amount of calculation is reduced more than half with small prediction accuracy loss.All kinds of parallel algorithms are implemented on CPU and GPU based on CUDA. The results of 1080p sequences’experiments show that compared with the serial algorithm, the accelerate ratio can reach 5.6 times without decreasing the image quality. The accelerate ratio of the fast intra prediction parallel algorithm can reach 8.5 times compared with the serial algorithm without decreasing the image quality.