Abstract:
Virtually all image compression techniques are designed to minimize the popular and easy-to-compute mean square error (MSE) measure of distortion. However, for images rep...Show MoreMetadata
Abstract:
Virtually all image compression techniques are designed to minimize the popular and easy-to-compute mean square error (MSE) measure of distortion. However, for images represented in the RGB color space the MSE is a poor match to the human visual system. Better results could be obtained by transforming the input image in a perceptually uniform color space such as the C.I.E. L∗u∗v∗, but such a color conversion is very time-consuming and for this reason it is rarely used. In this paper we propose to carry out the color space conversion by means of a high-rate hierarchical vector quantizer which requires only table lookup operations. As a consequence, the computational burned is reduced to a minimum and the transformation can be implemented efficiently both in hardware and software. Experimental results show that the the proposed fast color space transformation, although lossy, improves the visual quality of compressed images, as compared to that of images compressed without any color conversion.
Published in: 9th European Signal Processing Conference (EUSIPCO 1998)
Date of Conference: 08-11 September 1998
Date Added to IEEE Xplore: 23 April 2015
Print ISBN:978-960-7620-06-4
Conference Location: Rhodes, Greece