Remote Sensing Image Registration Based On Point Feature

Image registration is the process of aligning and overlaying two or more images of same target with different imaging conditions.Image registration is the basis of many processing technology,such as image change detection,3D reconstruction and target recognition,and it has been widely applied in many fields such as defense,satellite remote sensing,and pattern recognition.With the development of Satellite remote sensing technology,the methods of acquiring remote sensing images become more diverse.Owing to different conditions and imaging mechanism of sensors,gray change and noise distribution are of great difference in these images,image registration of such images is still facing many challenges.Therefore,the specific characteristics of these remote sensing images should be considered to explore appropriate registration method.According to the characteristics of remote sensing images,this paper improves the scale-invariant feature transform(SIFT)algorithm and synthetic aperture radar scale-invariant feature transform(SAR-SIFT)algorithm,and proposes the following three remote sensing image registration algorithm based on point feature:(1)Aiming at the problem that the noise and the significant gray changes lead to the failure of registration or poor matching result,a remote sensing image registration method based on phase congruency and spatial consistency is proposed.Then SIFT algorithm is used to detect feature points and generate the feature descriptors on the phase congruency images.By this way,the feature extraction methods in both spatial and frequency domains are combined.The experiments prove that phase congruency is robust to noise and can get rich texture information,and this method get enough effective features even there are great gray changes.Spatial consistency constraints are used for feature matching,and it improves the results of feature matching.(2)The neighbor information may be missed when gradient operator is used to extract features.And multiple main orientations of one keypoint will lead to interference in feature matching,only few matching points can be got.Aiming at these problems,a remote sensing image registration method based on fusion of edge and texture information and optimized Euclidean distance is proposed.By combining the texture information extracted by phase congruency and the edge information extracted by gradient operator,the feature descriptors are generated by using the fused information instead of the gradient.So most of the effective feature information will not be missed.Also the Euclidean distance is optimized for point matching,which effectively reduces the interference caused by multiple main orientations.It can be seen from the experiment results that our methods for feature extraction and feature matching are effective.(3)Aiming at the problem that keypoints with similar neighborhood information will lead to the mismatching and the omission of correct matching pairs.This paper proposes a remote sensing image registration method based on the regional constraint.Feature information and spatial area information are combined in feature matching.Based on the relative location relationship of correct matching pairs and the affine-invariance property of triangle-area representation,the regional constraint is constructed.By this way,the interference of points whose descriptors are similar but coordinate distance is of great difference is removed.The experiments prove that the number of mismatching decreases,and the number of correct matching pairs increases.