In the present paper, we address the problem of segmenting biological objects featuring corners. The main ingredients of our approach are automated feature-detection methods and mechanisms for introducing kinks in parametric spline snakes. We formulate a novel corner potential that enables the accurate segmentation of objects exhibiting sharp tips or acute angles. The optimization of active contours using the proposed keypoint-based energy yields robuster segmentation results and requires fewer parameters than traditional spline-snake approaches for the same task. The performance of our method is illustrated on microscopic images of two families of Rhabditidse roundworms.
Auke Ijspeert, Mohamed Bouri, Ali Reza Manzoori, Coline Lugaz, Tian Ye, Davide Malatesta
Simon Nessim Henein, Billy Nussbaumer