Extraction of Corresponding Points from Stereo Images by Using Intersections of Segments

In many applications, it is necessary to extract corresponding points automatically. In this study, we present a method of extracting corresponding points automatically from stereo images. First, we use the Combinatorial Hough Transform to extract feature points that are intersections of segments. This method is robust to the image noise. Then we correspond them by using backmatching method. Backmatching is robust to the occlusion. In conventional studies, mismatch is likely to occur. This is because they have used limited information in the image for extracting feature points. In our method, we use information of wide region in the image. Our proposed algorithm enables stable extraction of corresponding points in images composed of many edges, and this extraction is independent of minute threshold variations.

• Type: Conference
• Year: 2007
• Where: MVA
• Authors: Hiroshi Unno, Keikichi Hayashibe, Hitoshi Saji
• Keywords: Combinatorial Hough Transform, Computer Vision, Corresponding Points, Feature Points, MVA 2007