The Unmanned Aerial Vehicles Remote Sensing Image Registration Based On SIFT Algorithm

Image Registration is a fundamental problem in remote sensing, medical, computer vision, and many other fields. In the field of remote sensing, image registration is an essential step in realizing image fusion, change detection, image correction, image mosaic, and other applications. Because of the large remote sensing information and a wide range of applications, automatic registration has been a goal pursued by the people.Image feature extraction and image matching is an important method of automatic registration. In this paper, through analysis of a wide range of matching algorithm deeply, determine using the SIFT (Scale Invariant Feature Transform) as the ultimate method that used in the Unmanned Aerial Vehicles (UAV) remote sensing images point extraction. SIFT is a multi-scale feature extract method, it get great feature extract capability by construct DOG ( DOG scale-space , Difference of Gaussian scale-space) . At the same time, SIFT can generate SIFT features for describing the feature points. SIFT feature is invariable in the scale and rotation in the image changes; it also has a strong adaptability in illumination and image deformation. SIFT feature is useful in feature points match. Therefore, SIFT has a wide range of applications in the field of remote sensing, realizing image fusion, change detection, image correction, image mosaic, and other fields.SIFT algorithm mainly focused on feature information of the points. Geometric and Statistical information in matched feature-points are not concerned, so there are many mismatch feature points. In order to obtain more accurate matched feature points, moments and Mahalanobis distance were employed. Moments and Mahalanobis distance handle different object, so that, this is a multi-scale match algorithm. Finally, use new feature-points in image registration. Compared with the related work, our method can get more accurate and more matched-point. The experimental results demonstrate the robustness and efficiency of the algorithm.