Compression d'images par SVD et sur-approximation des composantes de chrominance

TL;DR

This paper introduces a novel color image compression method using singular value decomposition with over-approximation of chrominance components, achieving high-quality images at reduced data ratios and faster processing.

Contribution

It presents a new compression scheme combining chrominance sub-sampling and singular value over-estimation, improving speed and maintaining image quality.

Findings

High-quality image reconstruction at 15:1 compression ratio

Significant speed improvement due to sub-sampling

Effective over-approximation of chrominance singular values

Abstract

This paper gives a new scheme of colour image compression related to singular values matrix approximation. The image has to be converted in luminance / chrominance space before being processed like JPEG standard 4 : 2 : 0. Our approach is first based on a chrominance sub-sampling, then an over estimation of its singular values. Instead of keeping only the k first singular values for the 3 components R, G and B, we hold k first coefficients for the Y component and only k' (k' <= k) coefficients for 2 components Cb and Cr. Results show that for 512 x 512 pixels that, from k = 40 corresponding in an average distortion of 30 dB and a ratio of 15 : 1, the restored image has good quality. The algorithm allows a significant speed gain by sub-sampling too.