A practical consistent-quality two-pass VBR video coding algorithm for digital storage application

This paper presents a practical two-pass VBR coding algorithm for digital storage applications, which provides consistent visual quality perceptually and satisfies a fixed total bit-budget constraint. In the first-pass coding, we detect scene changes precisely and obtain scene complexity as well as other statistical parameters of the video sequence. The total target bits are allocated to frames optimally according to scene spatial and temporal complexity with decoder buffer overflow and underflow consideration. In the second-pass coding, we employ an improved iterative Qp selection algorithm to search for the optimal picture-level reference Qp that results in minimum difference between the number of coded bits and that of target bits. Adaptive quantization and the non-integer picture-level reference Qp selection guarantee uniform quantization artifacts perceptually and the precision of iterative search. Experimental results show that the proposed algorithm can offer consistent visual quality with smaller PSNR variation and higher average PSNR improvement as compared to TM5 algorithm and a typical two-pass coding algorithm. The application of the proposed algorithm includes DVD, digital library, VOD, and other digital media applications.