Signal Compression Based On Models of Human Perception

Nikil Jayant et al., Proc. IEEE 81(10), Oct. 1993

Summary | Relevance | Rating | Back to Index

Content Summary

An overview of the state of the art in perceptual image coding (compression based on elimination of redundant information as measured by human perception metrics). Introduces quantitative notion of perceptual entropy as the lowest bit rate possible for achieving perceived-lossless compression (just-noticeable distortion). Proposes a design space for perceptual coding algorithms whose axes are algorithm complexity, communication latency, bit rate and quality.

Discusses various "building blocks" of perceptual coding, including DCT, quantization, subband coding, VQ, wavelets and table-based compression, and shows how these interact in current compression schemes (e.g. JPEG). Video is still the area in which perceptual cues are least understood. Major research issues on the horizon are robustness, scalability, portability, and a unified theory of noise masking.

Relevance to Multimedia

Good survey of how perceptual-coding schemes are put together and why they are difficult to implement, and how perceptual metrics map to quantitative parameters of algorithmic building blocks.

Rating

3 out of 5: Valuable as a survey, but the "perceptual entropy" measure was not developed very much once introduced.

Armando Fox ([email protected])