Convex object contour estimation based on partially observed edges

Abstract

Segmentation and estimation of overlapping objects aim at automatic analysis of the visible edges to obtain the full contours of objects which can be of different shapes and have different degrees of overlap with each other making the task challenging. This research provides the review and comparison of different existing methods for the contour estimation that rely on the known visible contour points (evidences) and the constraint of convexity for the shape. Based on the comparison results, a new method based on the Gaussian Processes Regression was proposed improving the existing methods which use B-Splines and Ellipse Fitting. The experiments were carried out on a challenging dataset consisting of overlapping nanoparticles and showed the efficiency of the proposed method and its potential over the existing approaches.