Clump splitting via bottleneck detection and shape classification

Under-segmentation of an image with multiple objects is a common problem in image segmentation algorithms. This paper presents a novel approach for splitting clumps formed by multiple objects due to under-segmentation. The proposed algorithm includes three steps: (1) decide whether to split a candidate connected component by application-specific shape classification; (2) find a pair of points for clump splitting and (3) join the pair of selected points. In the first step, a shape classifier is applied to determine whether a connected component should be split. In the second step, a pair of points for splitting is detected using a bottleneck rule, under the assumption that the desired objects have roughly a convex shape. In the third step, the selected splitting points from step two are joined by finding the optimal splitting line between them, based on minimizing an image energy. The shape classifier is built offline via various shape features and a support vector machine. Steps two and three are application-independent. The performance of this method is evaluated using images from various applications. Experimental results show that the proposed approach outperforms the state-of-the-art algorithms for the clump splitting problem.

► We present a novel method for splitting clumps formed by multiple objects due to under-segmentation.
► We decide whether to split a candidate connected component by shape classification.
► We find a pair of points for splitting using a bottleneck rule.
► We join the selected splitting points by finding the optimal splitting line between them.
► We show that the proposed approach outperforms the state-of-the-art algorithms.