Research On Image Matching Algorithm Based On Improved SIFT Feature

Image matching refers to a method of analyzing image similarities by identifying feature points between two or more images by a certain matching algorithm.Scale In-variant Feature Transform(SIFT)is an important image matching algorithm based on local features.Because the algorithm has the advantage that the local features of the image remain unchanged during the rotation,scaling,translation,illumination effects and projection transformation,it has received extensive attention.The main work of this dissertation is to improve the feature points generation and matching methods in the traditional SIFT algorithm.Firstly,systematically explore the image feature extraction and matching methods in SIFT algorithm and Speeded Up Ro-bust Features(SURF)algorithm,and carry out experimental analysis and comparison.Then,it is found that the SIFT algorithm has some disadvantages such as high compu-tational complexity due to the high dimension of the feature points and easy mismatching due to a single matching condition.Finally,for these shortcomings,an improved method is proposed and an experimental analysis is carried out.as follows:(1)Since the feature vector generated by the traditional SIFT algorithm has high dimensionality,the calculation process is complicated and computationally intensive in the generation and matching phase of the feature vector.For this problem,this disserta-tion proposes an improved method to reduce the number of sub-pixel regions by re-dividing rectangular pixel regions,thereby reducing the feature vector dimension and reducing the time complexity of the algorithm;(2)In the feature vector matching stage,the traditional SIFT algorithm only matches by Euclidean distance.When there are multiple similar feature points,the mis-match problem is easy to occur.In order to solve this problem,this dissertation improves the matching conditions of the traditional SIFT algorithm.Different from the traditional SIFT algorithm,the SIFT algorithm in this dissertation incorporates the correlation co-efficient of the feature vector to filter the matching points that cannot be judged by the Euclidean distance,which reduces the mismatching point and reaches the goal of im-proving the matching accuracy.The experimental results show that the improved SIFT algorithm proposed in this dissertation is better than the traditional SIFT algorithm in terms of matching time and matching correctness.In the key point feature description sub-generation stage,since the dimension of the feature vector is reduced from 128 dimensions to 64 dimensions,the matching time is much better overall than the traditional SIFT algorithm.In the matching stage,due to the expansion of the constraint of the vector correlation coefficient,it sig-nificantly reduces the total number of repeated feature points,and largely eliminates the problem of many mismatches due to the large number of similar regions.Improved matching accuracy.Figure[31]table[4]reference[59]...