A Hough-like prediction/correction approach for shape extraction

Locating a two-dimensional object accurately in an image is a very common task in the applications of computer vision. In this paper, a Hough-like prediction/correction (HPC) approach is proposed for locating arbitrary shapes (i.e., non-analytical shapes) in images. The approach can be regarded as a modification of the generalized Hough transform (GHT) that has been shown as a robust method. Not only does the proposed approach retain the merits of the GHT, but also minimizes the problems.<
>