PROJECT TITLE :

Contrast Driven Elastica for Image Segmentation

ABSTRACT:

Minimization of boundary curvature is a classic regularization technique for image segmentation in the presence of noisy image data. Techniques for minimizing curvature have traditionally been derived from gradient descent strategies that could be trapped by a local minimum and, thus, needed a good initialization. Recently, combinatorial optimization techniques have overcome this barrier by providing solutions that may achieve a international optimum. But, curvature regularization methods can fail when the true object has high curvature. In these circumstances, existing methods rely on a knowledge term to beat the high curvature of the object. Unfortunately, the info term could be ambiguous in some pictures, which causes these methods conjointly to fail. To beat these issues, we tend to propose a contrast driven elastica model (together with curvature), that can accommodate high curvature objects and an ambiguous data model. We demonstrate that we tend to can accurately section very difficult synthetic and real pictures with ambiguous knowledge discrimination, poor boundary distinction, and sharp corners. We offer a quantitative evaluation of our segmentation approach when applied to a normal image segmentation knowledge set.