Abstract

This paper introduces a new error metric for binary images, defined as the p-th order mean difference between thresholded distance transforms of the two images. This has a theoretical justification related to topological ideas in mathematical morphology [14, 27, 39] and random set theory [24, 29, 45, 50]. It also has some intuitive interpretations. Theoretical development of the metric will be presented in a separate paper [2] (an earlier unsuccessful attempt was in [3]). The theory is also applicable to grey-level images, but here we describe only the implementation for binary images and compare it with standard measures such as the misclassification error rate and Pratt's [1, 31] figure of merit. 1

Citations

2367	A computational approach to edge detection - Canny - 1987
256	Distance transformations in digital images - Borgefors - 1986
148	On the statistical analysis of dirty pictures (with discussion - Besag - 1986
31	Quantitative design and evaluation of enhancement/ thresholding edge detectors - Abdou, Pratt - 1979
21	Convex Analysis and Measurable Multifunctions”, Lecture Notes in Math, no. 580 - Castaing, Valadier - 1977
11	Discussion of paper by - Besag - 1983
3	On convergence of closed sets in a metric space and distance functions - Beer - 1985
3	Metric spaces with nice closed balls and distance functions for closed sets - Beer - 1987
2	A class of image metrics - Baddeley - 1987
2	Range filters: intensity subrange filters and their properties - Bailey, Hodgson - 1985
1	Hausdorff metric for capacities - Baddeley