Researches On Representative Ground Target Matching Methods Based On Contour Features

According to the background of automatic target recognition methods, as well asthe specific application, this paper focuses on several important techniques ofrepresentative ground target matching recognition, including contour extraction, contourtracking and its simplification, different source image matching based on contourfeature. The three techniques researched in this paper construct a complete flow ofautomatic target recognition system, the main work and innovations are as follows:(1) Studying the contour extraction methods and proposing a new method based onsurround suppression. Firstly, the characteristics of infrared image and visible lightimage are analyzed, and the feasibility of using contour feature as the matching featurehas been proved by experiment. Secondly, a contour extraction method based onsurround suppression, which makes contour image more close to human vision sense, isproposed, and then its effectiveness of suppressing the texture edge has been proved byexperiment.(2) Researching on breaking points connection and simplification for the results ofcontour extraction and proposing a simplified algorithm based on improved D-Palgorithm. Firstly, several contour tracking methods are introduced, and breaking pointconnection method based on contour is selected because of its better simulation result.Secondly, the tracked contours are simplified and a simplified algorithm based onimproved D-P algorithm is proposed. The sampling points are found according to thecontours and whether these points are chosen as features is according to not only thecomparison of shifts and thresholds but also the self-crossing. Finally, the simplifiedcontinuous contours are achieved by connecting the feature points in order. Theeffectiveness of the improved algorithm is validated during the experiment.(3) Researching on the image matching methods of different source images, andproposing a method based on fold line. Firstly, the fold line description vector is built soas to find the corresponding fold line in different source images. Secondly, the area ofthe polygon which is constructed by the corresponding fold line is calculated and themathematic model between area and transformation parameters is made. Then theparameters are solved by minimizing the area. Finally, the matching result of differentsource image is achieved by resampling the image with the transformation parameters.Experiment results show the effectiveness and applications of proposed method.