Projective contour point matching using FPI, GRA and PSO

The contour point set is a very important feature for an object image. Finding the correspondences between two sets of contour points is a difficult task, especially under projective transformation. It has wide spread applications. For example, it can be used for object recognition by matching points derived from object models with points extracted from imagery. In this paper, a new contour point pattern matching (CPPM) algorithm using five-point invariant(FPI), grey relational analysis(GRA), and particle swarm optimization (PSO) is proposed. Firstly, two contour point sets from different images are extracted and normalized, then FPI is used to form the descriptors, GRA is employed to match the pair of point sets, and PSO is used to find exact corresponding pairs. Comparative experimental results manifest that the proposed method is more efficient, robust and fast than a comparative algorithm, RANdom SAmple Consensus(RANSAC) algorithm, for projective contour point sets matching.