Research On Image Matching Algorithm Based On Local Invariant Features

Image matching is an important machine vision processing technology.With the development of 3D reconstruction,medical image detection and image mosaic,the requirement of image matching technology is becoming higher and higher.The proposed algorithm is also being optimized,However,the disadvantages of high computational complexity and low matching accuracy still exist.In view of local invariant feature is the essential attribute of feature extraction of image content.And the extracted features are not affected by the specific forms of image content.The matching effect is very good in complex background interference and partial occlusion.So this paper studies image matching algorithm based on local invariant feature.Firstly,the method of image matching is studied.Based on the analysis of the advantages and disadvantages of various methods,the method of point feature extraction is selected.Then the point feature extraction method is introduced in detail and compared with experiment.Harris corner detection algorithm has unique advantages in the speed of operation and anti noise interference,especially the feature points extracted by Harris algorithm can show the characteristics of objects very well.SIFT algorithm has great advantages in scale transformation and rotation transformation.According to the advantages and disadvantages of Harris algorithm and SIFT algorithm,an improved Harris-SIFT algorithm is proposed.Firstly,in the aspect of feature detection,the Harris corner detection algorithm is added with scale parameters to construct multi-scale Harris corner detection operator.It can improve the adaptability to scale change based on the invariance of keeping rotation,illumination change and noise change.Secondly,in the aspect of feature description,the feature descriptor of SIFT algorithm is reduced by using the 28 dimension increasing homocentric square descriptor.This will greatly reduce the computation of matching algorithm on the basis of ensuring high matching rate.Finally,in the aspect of feature matching,the Euclidean distance is standardized and the standard Euclidean distance is used to measure the similarity.This will improve the matching accuracy of the algorithm on the basis of reducing the matching time.