Research On Image Feature Points Extraction And Matching Algorithm

The matching algorithm based on image feature points is an important research direction in the field of computer vision,and has caught widely attention by scholars at home and abroad in recent years.Compared with the traditional image matching algorithm based on gray value,image matching algorithm based on feature points matching has the higher operation efficiency and more stable effect.Therefore,in the field of target tracking,3D reconstruction and other practical engineering fields,the matching algorithm based on image feature points have been widely used.This paper mainly studies the extraction and description algorithm of image feature points and the optimization of feature space matching search strategy.Firstly,in the view of the fact that the traditional FAST image feature extraction algorithm has no invariance to the scale change of images,this paper proposes an improved FAST feature point extraction algorithm based on multi-scale space.The improved algorithm takes advantage of the spatial scale of SURF algorithm.In the multi-scale space of image,the traditional FAST feature points extraction algorithm is used to obtain the feature points at different scales,and the final feature points of images are obtained by means of non maximum suppression.Then,when the traditional FAST algorithm is used to extract the feature points,the detection threshold is usually single,which is determined by the experience before the experiment,and can not meet the requirements of extracting the feature points of different images.In view of the above two shortcomings of the traditional FAST image feature points extraction algorithm,in this paper,a modified multiple self-adaptive thresholds FAST feature points extraction algorithm base on image gray clustering is proposed.The algorithm uses the K-means clustering algorithm to classify the image,and in each cluster category,the maximum inter class variance algorithm is used to calculate the adaptive threshold of the category,in order to avoid the FAST algorithm because of the detection threshold constant shortcomings.Finally,because the dimension of the descriptor vector space is too high,the computation process of the matching search is very large.To solve these problems,this paper proposes an image feature space matching search strategy based on the teaching and learning optimization algorithm,and uses the teaching and learning optimization algorithm to simplify the search process of image feature space,so as to reduce the matching time.