The Research On Mixed Feature Extraction And Matching Algorithm For Large Affine Scene

The essence of the mixed feature extraction and matching algorithm for large affine scenes is to extract a variety of local stable invariant features.Since a single algorithm can not obtain many local stability features under the conditions of scaling,affine,illumination,rotation transformation etc.,it is necessary to merge the algorithms which can extract the stable features,so as to obtain more stable matching point.The maximum stability extreme region algorithm based on grid division is merged with the affine scale invariant feature transformation algorithm.Because of the matching parameters of region feature are used as the constraint condition of point feature matching,a large number of correct matching points are obtained.Firstly,it is described the aperture imaging model and the related theory of two-dimensional geometric transformation of image,and emphatically elaborates the principle of affine transformation and the invariant property of affine.It is introduced the related algorithms of local invariant feature extraction and matching.In order to solve the problem mentioned above,it is proved that the general feature extraction and matching algorithm can not solve these problems,and analyzed the experimental results of these algorithms.Based on the idea of algorithm fusion in this paper,it is introduced the algorithm of the maximum stability extreme region,and it has been improved too.It is used to judge the richness of the regional information by the information entropy of the extreme region on the condition of dividing the image into regions,and then removed the extreme regions with low abundance.The regional features are evenly distributed in the image,and finally the results of the algorithm are compared and analyzed.Then,it is focused on the feature extraction and matching algorithm of multi-feature fusion.It extracts the stable features of irregular regions,fits the ellipses.After the feature selection through the grid division,it is obtained that the geometric relationships between the correct matching feature pairs which as the constraints of the point feature matching.Then extracted the point feature,removed the mismatches according to the constraints and then mixed with the correct matching pairs of regional features.It can obtain a large number of correct matching pairs.The fusion algorithm was tested making use of multiple sets of images,and the results were analyzed.Finally,it is described the CUDA's parallel optimization technique.The feature extraction and matching algorithm of multi-feature fusion is optimized in parallel and the optimization results are compared.It shows that the parallel optimization can greatly improve the efficiency of the algorithm without affecting the accuracy of the results by experimental results.