A Modified Segmentation Approach for Overlapping Elliptical Objects with Various Sizes

Segmentation of elliptical objects has many real-world applications including morphology analysis on biological cell, material particles and other objects which need quantitative analysis according to size and shape. However, overlapping and varying in size may make the objects segmentation extremely challenging. In this paper, a modified segmentation approach for overlapping objects with different sizes is proposed. Specifically, we extract all the concave points for each connected region from the object’s silhouette. We next fit all of the circles by two adjacent concave points and an arbitrary point which is on the edge right between the two concave points. A radius set is extracted from all the circles, and a segments set is determined by the edge fragments between all the two adjacent concave points. Based on the radius set and the segments set, we can determine if there is a large gap in the radius set and the length of segments corresponding to the radius. The edge segments and radius set are divided into two subsets, while the appropriate radius range and threshold are selected respectively from the two subsets to execute the Bounded Erosion-Fast Radial Symmetry transform to get the seed point for each object. Our experiments are taken under synthetic and real datasets, in which the overlapping objects in these datasets are with different size. The experimental outcomes show that the proposed approach outperforms other existing schemes.