Efficient Rate-Control System With Three Stages for JPEG2000 Image Coding

Abstract

Although the conventional post-compression rate distortion (PCRD) optimum algorithm in JPEG2000 can provide excellent rate-distortion performance, it remains troublesome for VLSI design. It requires high-complexity computation and a large quantity of working memory. In this paper, a novel rate-control system for JPEG2000 image coding, which consists of three stages, is proposed to efficiently reduce the computational complexity and memory usage, as well as maintaining the high image comparable with PCRD. In the precoding stage, the rate allocation for the image tiles (RAIT) can allocate the bit rates for different image tiles based on their internal information; moreover, in order to make VLSI design easier, the image edge-detect algorithms and the prediction algorithms are used to evaluate the information of the image tiles. In the tier1 coding stage, according to the given bit rate, the rate allocation based on the rest rate distortion (RARRD) can adaptively truncate code stream based on the rest rate distortion of the current image tile and the rate-distortion (RD) slope of the current coding pass during the current code-block coding, and all of the truncated data need not be encoded and stored. In the post-coding stage, a new fast rate control with approximately optimum truncation (RCAOT) can get the approximately optimum truncation points through searching among the 256 quality layers once, which are divided based on the 8-b RD-slopes expressed by means of a logarithm. This system can not only maintain high image quality through RCAOT and RAIT but also save working memory and encoding time by means of RARRD and RCAOT. Experimental results show that, for the image which has distinct objects and background, the PSNR of the reconstructed images compressed by our rate-control system are much higher than those using PCRD; for other images, our system has the comparative effect as PCRD