ROBUST TECHNIQUES FOR EDGE-DETECTION IN MULTIPLICATIVE WEIBULL IMAGE NOISE

Recent studies of the approximately multiplicative noise present in images from coherent radar systems suggest that the first-order statistics can be broadly modeled by the two-parameter Weibull probability density function w(y) = αβyβ-1 exp(-αyβ). This paper defines and compares some novel techniques for the robust detection of sustained image irradiance changes, or edges, in images immersed in multiplicative Weibull noise. It is shown that edge detection in multiplicative noise is accomplished effectively by comparing ratios of locally adjacent image irradiance estimates to an appropriate threshold. Several such estimates are considered: local irradiance averages, optimal single order statistics, and best linear unbiased estimates using order statistics. The results of a statistical (Monte Carlo) performance analysis suggest that edge detection using ratios of single order statistics (ROS detector) offers the best compromise among computational convenience, edge localization and robust performance, while the ratio of BLUEs detector (ROB detector) yields the best overall statistical and empirical noise suppression. The ratio-of-averages detector (ROA detector), while statistically optimal in some cases, often yields inferior edge localization relative to the other schemes. Images corrupted by synthetically-generated Weibull noise are used to support the analysis. © 1990.