Department of Electrical and Computer Engineering,
Engineering Science Building,
The University of Texas at Austin,
Austin, TX 78712-1084
arslan@ece.utexas.edu -
magesh@ece.utexas.edu -
bevans@ece.utexas.edu
Synthetic aperture radar (SAR) systems are mounted on airplanes and satellites, which have limited downlink and storage capacity, yet SAR image sequences may be produced at rates of several Gbps. Compression is difficult because SAR images contain significant high-frequency information, such as terrain boundaries and terrain texture. In assessing the quality of compressed images, peak signal-to-noise ratio and mean-squared error are inadequate because they assume that distortion is solely due to image-independent additive noise. In this paper, we provide objective measures to assess the visual quality of SAR images compressed by JPEG and SPIHT coders. The human visual system responds differently to linear distortion and noise injection (nonlinear distortion plus additive noise). Our key contributions are that we (1) decouple and quantify the linear distortion and noise injection in JPEG and SPIHT coders, and (2) introduce a new edge correlation quality measure which we use to quantify nonlinear distortion.
Last Updated 11/13/99.