Invariant line segmentation for object recognition

A fast method of line segmentation using the dominant point approach is described. It is based on Lowe's method. The contribution of this paper is successfully finding the way of selecting initial points for starting Lowe's algorithms. This is a crucial step for getting dominant points which are stable under variant scales and orientations. An one-dimensional pyramid structure is designed to speed up the process. In addition, a parallel algorithm for edge linking is also illustrated. This further reduces the overall speed for line grouping.<
>