Research On Image Matching Based On Improved SIFT Algorithm

With the development of modern science and technology and the progress of computer technology,the research of image matching is becoming more and more in-depth,and image matching has become a very important technology in image processing and computer vision.The image matching is the process of comparing the images collected in different time and different angles.We can find out the similarities in two or more images by image matching.At present,the image matching technology has been widely used in the fields of biomedicine,aerospace,virtual reality and so on.In order to reduce the influence of these error factors on the image matching performance,the image is needed to filter,image enhancement and transform before image matching,because the image is affected by the vibration of the acquisition device,the impulse before the camera and so on.In the case of an image matching method,it is usually judged by the three aspects of matching speed,matching accuracy and stability to determine its' good or bad.The SIFT algorithm is a kind of classical image matching algorithm,which is widely used,and the SIFT algorithm has transformed the SURF algorithm,the PCA-SIFT algorithm,etc.Although the operation of SIFT algorithm is very stable,the dimension of its characteristic descriptor is too high,which inevitably leads to the problem of slow operation and low efficiency.These two deformation algorithms of SIFT algorithm have improved the efficiency of the algorithm because of the change of the feature extraction method and the feature descriptor,but the matching accuracy is not as good as the SIFT algorithm.In order to solve the above problems,this paper proposes a hierarchical radial partition method to construct the feature descriptor.The neighborhood of the feature point is divided into 8 regions,the 64-dimensional feature descriptor is obtained by counting the histogram of the gradient direction in 8 directions in each region,So that the feature description sub-dimension is reduced by 50%.At the same time,because the Mahalanobis distance considering the correlation between the sub-vectors of the characterization descriptor,so,the Mahalanobis distance bidirectional matching method is used to replace the Euclidean distance in matching,and the RANSAC method is used to eliminate the mismatch point.The experimental results show that the improved SIFT algorithm preserves the advantages of the original algorithm stability and increases the matching speed greatly,and enhances matching accuracy.Finally,the improved SIFT algorithm is applied to the matching of the car rearview mirror switch image,and achieves good results.