Machine Learning Based Image Matching Methods Research

With the rapid development of intelligent manufacturing,machine vision has become one of the indispensable technology in various fields.With the help of image analysing and processing capabilities by computer,human vision guiding simulation and mechanical action control are bound to be intelligent manufacturing trend.Image Matching technology,a branch in machine vision,is a key step of many image processing technologies.The overall performances of the system are always determined by image matching accuracy.However,existing matching technologies are unable to meet the growing new demand.A new theory is badly in need.The work of this paper in image matching technology is as follows:1)Review the existing image matching techniques,and analyze the purpose and significance of research Image Matching technology.Then,according to the two categories based on feature and pixel gray scale,summarize the current domestic and foreign research respectively.All those lay the theoretical foundation for the following research.2)Research on pixel gray matching technology.By the introduction of the common matching algorithms and operators,we introduce "Lucas-Kanade algorithm",a widely used technology in many areas.Those will be studied in depth and mathematical derivation will be conducted.3)Research on feature-based image matching technique.We introduce several classic feature extraction algorithms,and analyze an edge feature based image matching algorithm.This algorithm is tested to be robust to complex scenes,occlusions and illumination changes.4)Research on point pattern matching algorithms.By detailed study on current point pattern matching algorithms,we proposed an genetic algorithm based on non-rigid point matching algorithm then effectiveness is approved by experiments.The above two image matching techniques provide direct and effective reference to the first-line engineers.The point matching algorithm based on genetic algorithm not only shows perfect performance,but also expends the application field of genetic algorithms.All those provide reference for the combinations of intelligent learning method and the traditional matching algorithm.