Image compression with singularity preservation

Abstract

This paper presents a new approach to lossy image compression through singularity preservation to obtain high compression ratios. The concept of this approach is based on a conjecture that image singularities carry most of the perceptual information, hence the essential part of an image should be represented by its singularity as opposed to its energy alone. Wavelet maxima have been chosen to represent signal singularity because of their ability to characterize image singularity fully. There are algorithms to reconstruct the original image faithfully from wavelet maxima. A compression scheme can then be designed to reduce the bit rate while preserving singularities. The resulting low bit-rate image has sharp edges without distortions, such as blockiness or blurs. This approach has been used to compress aerial ortho images, in which the perceptual quality of a 27 peak signal-to-noise ratio (PSNR) singularity-preserving image outperforms that of a 30 dB PSNR energy-preserving joint-photographic expert group (JPEG) image at a 15:1 compression ratio.