Greedy perceptual coding

This paper discusses an approach to quantization for lossy compression, termed greedy perceptual coding, in which the quantizer uses a perceptual model to determine the amount by which a work of media can be distorted, and then greedily tries to find the distortion that minimizes the number of bits that will be output by a subsequent lossless coder. The chief advantage of this approach is that the decoder does not need to know any distortion parameters used during compression, such as quantization step sizes or perceptual model parameters. As a result, the perceptual model can be arbitrarily complex. Since greedy perceptual coding changes the distribution of image values to be encoded, the best lossless codes for undistorted images are not the best to use in this context. A method is presented here for designing simple codes to use in the greedy-perceptual-coding context, and a preliminary compression system is implemented using these ideas