Research On Object Matching Algorithm Based On Contour Feature

Contour as a high level visual features, which has a stable description of characteristics of the target. It is widely used in artificial intelligence, high-end equipment, satellite remote sensing, medical imaging and other fields, it has been becoming the hot topic in the field of pattern recognition. Three aspects of based on extracting contour feature, generating feature descriptors and measuring similarity measure are studied in this paper, the problems of detecting the TFDS train handle and positioning the dust cleaner and FPC fill strong pieces were resolved successfully. The main work of this paper is as follows:The image algorithms of preprocessing and contour extraction are analyzed, lots of contrast experiments which combined with a specific matching objects(handle, FPC fill strong piece and dust collector) were conducted. Finally, the optimal schemes of each of the matched object is established.The feature definition, similarity measure and matching strategy of contour matching algorithm are deeply studied and the concrete application forms of various measures are discussed. Also we analyzing the constructions of geometric primitives descriptors, the outer contour rectangle, circle, convex, polygon of contour are discussed respectively. So the problem of FPC fill strong pieces position is resolved successfully.A variable step matching method based Hu contour invariant moments and Gaussian pyramid model is proposed. the weight step formula is designed according to the shape of the target and program running time is reduced greatly combining with the Gaussian pyramid sampling model and termination threshold. Firstly, the method is begun to locate optimum matching position in original image layer through matching layer mapping of Gaussian pyramid model, besides, the termination of the threshold suppression is formulated to curtail similarity matching which is farther and the matching search path is developed based on weights of transversal matching variable step size, finally locking region where the handle lies in. the matching speed improves greatly, obtaining good matching performance.A shape matching method based on geometric features is proposed. Firstly, the contour points are sampled and initial location of critical points and mapping relation of point set are determined basing on polar radius and local curvature. Then taking centroid as a benchmark, dual descriptors with geometric features of angle and scale are generated and processed by the standard quantification. Finally, the similarity of the descriptors is calculated with modified Manhattan distance. The experimental results have shown that the proposed shape matching method is hardly influenced by geometric transformation such as expansion, rotation, and translation, which demonstrates its adaptability and robust.The analysis and comparison between Hu contour invariant moments and geometry dual descriptor matching algorithms is completed, summing up the general conclusions on the matching effect and time consumption.Experimental results show that the algorithm has a significant advantage compared with other common matching algorithms, and it provides a new way of thinking for contour matching algorithm.