Segmentation and grouping of object boundaries

A method is presented to detect curvilinear structures in 2D images. The approach is based on a generalization of the Hough transform that is different from its traditional, template-based generalizations. With this method, global interaction of edge elements is reduced through the use of the Delaunay graph in utilizing the Hough transform. The algorithm as it is currently implemented needs to be extended. One suggested extension is to perform a connected component analysis of the groups obtained from current algorithm and use the distribution of the set of bins in Hough space to further group these larger components. Another extension would be to use the Hough space information to perform contour completion for occluded objects.<
>