Scene Matching Algorithm, Performance Evaluation And Its Application

Scene matching is an important image analysis and processing technique, which located a image area from the other image or find out the correspondence between them, where the images taken at different times, by the different sensors, or by different viewpoints from the same scene. Scene matching technique is used in a wide-range areas such as navigation and location, object recognition, movement analysis, stereo vision, data fusion, change detection. It is the base of these image processing techniques. The difficulty of designing matching algorithm lies in the variety of scene styles and variety of imaging variations between images. How to design an algorithm with high adaptability, high precision and computing efficiency is the kernel study topic. At currently, problems of matching navigation and sensor fusion on cruise missile need to be settled urgently, In this paper, scene matching technique is studied from viewpoints of feature-based matching algorithm, algorithm performance evaluation and matching area selection, scene matching vision expert system.In the first part of the paper, scene matching problem as a whole is analysed and discussed. The mathematics description of scene matching problem, factors that affect matching performance, framework to design scene matching algorithm including feature space, similarity merits, search space and search algorithms are detailed sequentially. The ideas to divide the variation into four styles and divide a matching algorithm into four factors are the director to design a matching algorithm.In the second part of the paper, feature-based algorithms are studied. The algorithms include edge-based correlation matching algorithm, edge-based hausdorrf distance matching algorithm, area moment based algorithm, point based algorithm and graph based algorithm. The feature extraction, similarity merit, search algorithm are detailed. Their performance on adaptability, precision and computing efficiency are analysed and compared. The experiment results are also given.In the third part of the paper, algorithm performance evaluation and matching area selection is studied. The system architecture and implement of matching algorithm performance evaluation through statistic experiments is developed. And then through the study of relations between image characteristic, imaging variation and matching performance, the method and criteria to select matching area is proposed.In the fourth part of the paper, the study of scene matching vision expert system integrates the study of scene matching algorithms and performance evaluation into a design style vision expert system. Agent technique is selected and multi-agents cooperate to resolve a scene match problem. The system architecture is designed and their workflow is explained.Finally, the author's idea to extent study is given in the chapter of tag.