Nonparametric tests for edge detection in noise

摘要

In this paper we describe three statistically motivated techniques for detecting edges in gray-level images. Two similar methods based on linear rank sums are described; specifically, the Wilcoxon and median statistics are implemented in a modified form and are found to perform effectively on both noisy and uncontaminated sample images. A novel approach based on fitting order statistics to the image data is also advanced. Numerous examples are given and computational requirements are examined.