Determining straight line correspondences from intensity images

摘要

This paper presents an algorithm for straight edge extraction from intensity images and an algorithm for straight line matching. In the first part of the paper, straight edge extraction is described. Image data are first processed by the operation of edge support focusing to remove unnecessary image details. An edge support is formed in this process. Then, straight edges are extracted from line support regions, which are segmented from the edge support. In the second part of this paper, we describe straight line matching using a matching function, which characterizes the similarity of edge lines of two images and is based on not only the geometrical relations of the lines but also the information from the intensity images. A technique of two-step matching is applied to reduce the cost of computation.The motivation behind this paper is our work on motion estimation from line correspondences of sequential images where straight edge extraction and matching is an essential first step. The output data of the algorithm can also be used for calibration of stereo camera setups. The results of experiments using indoor and outdoor scene images are shown.