Research On Image Registration Technique Based On Features

As one of the core technologies in the field of digital image,image registration has been widely applied to remote sensing detection,medical research and military fields.However,there are many difficulties in the process of image registration,such as illumination brightness change,scale change,affine transformation,similar object interference and so on.With the development of technology,a variety of efficient registration algorithms have been proposed,in which the method of image registration based on feature points is stabler and more suitable at present.The algorithm based on the feature points can be classified into feature points based on floating-point descriptors and based on the binary-code descriptors.The most classic algorithm for floating-point descriptors is scale-invariant feature transform(SIFT).The registration algorithm based on SIFT feature points can maintain relatively good robustness under different conditions such as image rotation,zoom in and out,translation and so on.In this paper,SIFT registration algorithm is deeply studied and the following improved algorithm is proposed.When extracting feature points for SIFT algorithm,a large number of unstable edge points and useless points are generated,a new feature point extraction method is proposed to solve the problem in this paper.Firstly,Harris corner points are applied to gaussian image pyramid to make the scale diversity of Harris corner points extraction.On this basis,using the Canny edge operator to extract the characteristics of the extraction,so as to make up for the problem of the extraction in the weak marginal region of the Harris point.Then,the feature points that combine Harris corner with Canny edge feature are characterized to continue the image registration.Experimental results show that the feature points extracted by the improved algorithm have mostly meaningful points,which can not only reduce SIFT useless points,but also reduce registration time and improve registration accuracy.After image matching,some false matching point pairs still exist due to the influence of image noise and affine transformation,which result in inaccurate registration result.Based on this phenomenon,an improved false matching point removal technique is proposed in this paper.In the iterative process of calculating the transformation model by the traditional random sampling consistency(RANSAC)algorithm,the mismatching points are also iteratively iterated,thereby increasing the registration time and reducing the registrationefficiency.The paper introduces the root mean square error(RMSE)to delete the error matching points generated during the RANSAN iteration process in advance to reduce the iteration time and improve the registration efficiency.The experimental results show that the improved algorithm can reduce the image matching time and improve the registration efficiency.To verify the stability and efficiency of this algorithm,the image registration technology based on improved feature points and improved RANSAC algorithm is applied to high-resolution images,remote sensing images and natural image scenes.Experimental results show that compared with other excellent algorithms,the improved algorithm can obtain better registration images with fewer iterations,and has good usability and high efficiency.