Elliptical Distance Transforms And Object Splitting
Hugues Talbot, Ben Appleton · 2002
The classical morphological method to separate fused objects in binary images is to use the watershed transform on the complement of the distance transform of the binary image. This method assumes roughly disk-like objects and cannot separate objects when they are fused together beyond a certain point. In this paper we revisit the issue by assuming that fused objects are unions of ellipses rather than mere disks. The problem is recast in terms of finding the constituent primary grains given a boolean model of ellipses. To this end, we modify the well-known pseudo-Euclidean distance transform algorithm to generate arbitrary elliptical distance transforms to reduce the dimension of the problem and we present a goodness-of-fit measure that allows us to select ellipses. The results of the methods are given on both synthetic sample boolean models and real data.