Abstract:

Point pattern matching (PPM) is an important issue in computer vision and pattern recognition, such as target recognition, medical and remote image registration, and pose estimation, and so on. We propose a novel PPM algorithm based on point features. We first construct feature attribute image of the points according to the distribution of point sets and points’ position. Then polar coordinate transform is applied to the feature attribute image, and we describe the feature attribute image with the moment invariants. We get the course matching results by comparing the feature vectors, and an iterative method is introduced for the final matching. We make two contributions in this paper. Firstly, we construct a polar coordinate transform based points  features, and describe the features with the moment invariants. The features are invariant to rotation and transportation. Secondly, an iterative matching process by point  features and geometry constraint is proposed. The method is insensitive to outliers and noises. Experimental results demonstrate the validity and robustness of the algorithm.

Key words: computer vision;point pattern matching;polar coordinate feature image;moment invariants