Efficient, ‘greedy’ rate allocation for JPEG2000

Abstract

We describe an efficient algorithm for rate control in JPEG2000 or similar sub-band based visual compression schemes. Operating in tandem with the compression process, our method initially predicts “Distortion-Rate” (RD) slopes for all code-blocks' coding data, based on empirical statistics of coefficients and subsequently uses a greedy approach. We use an optimized doubly-linked tree structure, which is initially constructed using predicted slopes as the key. The tree is subsequently updated with actual slopes to proceed with our rate allocation method. Our algorithm allows for fine-grain targeting of bitrates. A good balance of complexity and quality is observed in this scheme vis-à-vis the full effort Post Compression Rate Distortion (PCRD) optimization. Quality is optimized across a multitude of code-blocks competing for their constituent coding passes to be embedded into the bitstream, with good `dispersion' of distortion.