H.264視訊編碼之PSNR品質提升與可適性並行排程之設計

Abstract

H.264 is one of the latest block-oriented motion estimation based video coding standards. It provides much better image quality and compression ratio with the large increment of computational complexity compared to previous coding standards. How to take the balance between coding quality and efficiency becomes an important issue. In this thesis, we target on the characteristic of parallelism of H.264 encoder for achieving higher throughput with the architecture of heterogeneous multi-core SoCs and design a PSNR-based algorithm which takes care of the image coding quality. Since the high complexity of motion estimation, it adopts function partition parallelism and separates the encoder from motion estimation module and other function modules. The PSNR-based algorithm controls the parameters which affect the image coding quality with the pipelined scheduling for H.264 encoder. With these mechanisms, we enhance the coding throughput by pipelined design with the architecture of heterogeneous multi-core system and keeps the coding quality with the PSNR-based algorithm. When compared to other fast algorithms for H.264 encoder, the result shows that the PSNR-based algorithm with pipelined scheduling achieves about 1.6 times of speed up in coding throughput and keeps the coding quality.