Research On Target Recognition Based On Improved SURF Algorithm

As part of a modern target recognition technology, Image-based object recognition technology has become a focus of research in the field of imaging applications. Image-based object recognition is a process that using a two-dimensional image or a video sequence finds objects. The process includes two steps: Firstly, building a collection of feature vectors by extracting the target image features and making this collection of feature vectors as the basis for target recognition. Secondly, matching the target feature vectors and the target identified feature vectors obtains and analyzes the results of matches. Completing the target recognition by using recognition and judgment algorithm. Because using the local characteristics of the target image describes the target object characterize with a certain degree of stability and good uniqueness. So the researches for target recognition have certain theoretical basis and practical significance.The thesis mainly does some researches about the process of Image-based object recognition technology. The main work is as follows:Firstly, Describing several image feature extraction algorithm and describing SURF algorithm in detail is to propose an improved algorithm to extract the feature information of image. And making the corresponding theoretical analysis and experiments.Secondly, Introducing several feature matching algorithm based on feature vectors and making the corresponding theoretical analysis and experiments is to choose KD-Tree algorithm to match between the feature vectors. By analyzing the eliminating false matching methods in the literature,presenting a method based on position error to eliminate false matching feature points and making the theoretical analysis and experiments.Thirdly, building the framework of target recognition system based on local feature and Programming this system that consist of image displaying system, extraction system, target image registration system and target recognition system and making the theoretical analysis and experiments for every subsystem .Integrating every subsystem is to build target recognition software. Improved SURF algorithm is used to do some experiments that include target rotation, target scaling and target sheltering by the target recognition software, and Theoretical and experimental results were analyzed and summarized.Finally, conclusions are given with recommendation for future work.