Color VQ-based Image Compression by Manifold Learning

Christophe Charrier (1, 2) , Olivier Lezoray (2)

1 MOIVRE Centre
2 Equipe Image - Laboratoire GREYC - UMR6072

When the amount of color data is reduced in a lossy compression scheme, the question of the use of a color distance is crucial, since no total order exists in IRn, n > 1. Yet, all existing color distance formulae have severe application limitation, even if they are widely used, and not necesseraly within the initial context they have been developed for. In this paper, a manifold learning approach is applied to reduce the dimension of data in a Vector Quantization approach to obtain data expressed in IR. Three different techniques are applied before construct the codebook. Comparaisons with the standard LBG-based VQ method are performed to judge the performance of the proposed approach using PSNR, MS-SSIM and VSNR measures.

Domaines: Traitement des images [eess.IV]

Christophe Charrier, Olivier Lezoray. Color VQ-based Image Compression by Manifold Learning. International Conference on Image and Signal Processing, 2010, Trois-Rivières, Canada. pp.79-85, ⟨10.1007/978-3-642-13681-8_10⟩. ⟨hal-00521069⟩