Model-based Edge Tracking for Segmentation of Low Contrast Images

Abstract

Segmentation is a significant preliminary step for many image-based object recognition activities. Microscopy images often present segmentation problems, namely low contrast (the objects are translucent) and occlusions. Fortunately, translucency provides some possibility of solving the occlusion problem; edge-based methods can be used to tackle the low contrast (translucency) problem, but the edges are noisy and edge tracking must be used. In occluded regions edges can be very faint and noise and conflicting edges can confuse even edge tracking: an edge contour containing gaps may result. This poster presents work on a gap filling algorithm that uses model-based prediction to augment noisy edge data.