4 Collinear Points: Robust Point Set Registration Using Cross Ratio Invariance

In this paper, we present a robust point set registration method based on the cross ratio invariance of 4 collinear points, which is able to deal with point set registration problem under the most general linear transformation, projective transformation. On the basis of all combinations of 4 collinear points extracted by Hough transform, meaningful correspondences are identified by combining the cross ratio invariance of 4 collinear points and Randomized RANSAC. At the end, the underlying projective transformation matrix was estimate in a least square sense. It has been shown in the simulation experiments that the proposed approach remains robust in high level of degradations, including 45% outliers and 40% overlap ratio. Experiments with Oxford corridor sequence and ZuBuD wide baseline image database proved its usefulness in real application.