Texture compression with adaptive block partitions (poster session)

  • Authors:
  • Leonid Levkovich-Maslyuk;Pavel Kalyuzhny;Alexander Zhirkov

  • Affiliations:
  • The Keldysh Inst. of Appl. Math., 4, Miusskaya Sq., Moscow, Russia, 125047;Lomonosov Moscow State University Russia, Moscow, GSP119899, MSU, Computer Science Faculty;Lomonosov Moscow State University Russia, Moscow, GSP119899, MSU, Computer Science Faculty

  • Venue:
  • MULTIMEDIA '00 Proceedings of the eighth ACM international conference on Multimedia
  • Year:
  • 2000

Quantified Score

Hi-index 0.00

Visualization

Abstract

We present image compression method based on block palletizing. Image block is partitioned into four subsets, and each subset is palletized by 2 or 4 colors from the quasioptimal local palette, constructed for the whole block. Index map for the whole block, being the union of index maps for subsets, is thus only 1 or 2 bits deep, while the local palette may consist of 8 or even 16 colors. The local palette has a specific geometrical configuration in RGB color space, determined by only 2 colors. These two colors are stored explicitly, and the rest are reconstructed at the decompression stage. Compressed block consists, essentially, of the index map, palette description and partition description. This format allows fast access to randomly chosen pixels, and high reconstruction quality for compression ratios from 8 to 12, which is useful for texture storage in 3D graphics applications where real-time decompression is crucial.