Uniform Quantization Error of Laplacian Sources with Applications to JPEG Coding

J.Minguillón and J.Pujol

Abstract. In this paper we propose a novel method for computing JPEG quantization matrices based on desired mean square error, avoiding the classical trial and error procedure. First, we find a relationship between a Laplacian source and its quantization error when uniform quantization is used. Then we apply this model to the coefficients obtained in the JPEG standard once the image to be compressed has been transformed by the discrete cosine transform. This allows us to compress an image using JPEG standard under a global MSE (or PSNR) constraint and a set of local constraints determined by JPEG standard and visual criteria. Simulations show that our method yields better quantization matrices than the classical method of scaling the JPEG default quantization matrix, with a cost lower than coding, decoding and error measuring procedure.

Key words. Laplacian sources, uniform quantization, JPEG

Registration: PIRDI-2/98, February 1998.

Retrieve PostScript document (pirdi2.ps: 520698 bytes)
Retrieve PostScript gzipped document (pirdi2.ps.gz: 110058 bytes)
Retrieve DVI document ( pirdi2.dvi: 120084 bytes)
Copyright © 1998 by Computer Science Department. UAB