2D shape reconstruction from the Hough transform

A new algorithm uses information available in the distribution of peaks in parameter space to calculate the coordinates of vertices in the image, thereby eliminating the need to refer back to the image to reconstruct it. The method performs best for shapes that are closed, as many real objects are. Two methods are proposed to determine the end-point information. One follows the work of Chan (1992) where identification of peak colinearities in the transform space is used to locate vertices in polyhedral images. The second approach tracks the edge-point information used to update all those accumulator cells which have more votes than a set threshold. When the transform is completed, the starting points of the lines present in the image are calculated. By integration of the two methods, a new algorithm uses the information obtained by both to reconstruct the image