A Mixed Parallel and Pipelined Efficient Architecture for Intra Prediction Scheme in HEVC

Abstract

The complexity of intra prediction in High-Efficiency Video Coding (HEVC) is increased significantly due to the incorporation of inherent features like variable-sized quadtree partitioned coding units and 35 angular modes that help in achieving better compression. This paper presents an efficient hardware architecture for the intra prediction that supports and comprises the above aspects and achieves a higher throughput to support high definition (HD) videos. A compact reusable reference buffer structure is implemented to limit the buffer size to 1 KB. A dedicated arithmetic unit to take advantage of the parallelism present in the prediction algorithm is incorporated, which allows the reuse of multipliers to reduce hardware resources. The loading of reference samples to buffers for prediction causes significant delays which are eliminated in our design. The entire architecture functions as a pipelined unit with no data dependency and generates eight samples/clock cycle in parallel. The design is implemented on a Field Programmable Gate Array (FPGA) platform operating at a frequency of 110 MHz. This makes it possible to support 4 K videos at 30 frames per second, with the resource cost of 16 K logic gates and 122 registers.