Citation

Abstract

Distortion-controlled data compression is lossy signal compression in which the amount of distortion introduced into any small portion of a signal is strictly limited. This article gives results on various practical aspects of quantization and distortioncontrolled compression. The primary focus is near-lossless compression, as accomplished by predictive techniques. For compression of noisy signals, the effect of quantization step size q on rate, distortion, and the effective noise characteristics is considered. It is demonstrated that surprisingly large values of q often may be employed without significantly affecting the scientific analysis of the compressed data. For low-noise signals, the use of subtractive dither to reduce or eliminate compression artifacts is analyzed. This analysis includes determination of a class of optimal dither signals based on certain reasonable assumptions. The effect on compression of the step size used in analog-to-digital conversion also is discussed.

Keywords

quantization near-lossless compression image compression dither distortion-controlled compression predictive compression compression artifacts

Details

Volume
42-139
Published
November 15, 1999
Pages
1–38
File Size
901.3 KB