Research On Registration Of Infrared Multi-spectral Images With Large Field Of View Based On Points And Lines Features

The large field of view infrared multi-spectral scanner is specially developed by the national high-resolution earth observation system.The scanner generates 10 bands of image data,which can realize multi-spectral,large field of view and high-resolution earth observation.In order to quickly obtain 10 bands of spatially aligned infrared multi-spectral products and further improve the geographic location accuracy of mosaic infrared image products,it is necessary to accurately register the images.However,there are many differences between the images,making the traditional registration methods fails to obtain ideal results.Therefore,this paper proposes a new image registration method adapts to these images.The main contents of this paper are as follows:1.A phase congruency-based corner screening and feature matching method is proposed in this paper.Firstly,feature points are extracted and selected using the minimum moment and the maximum moment of phase congruency.And then,feature descriptors of the retained feature points are constructed by using both the magnitude and orientation of phase congruency.Initial matches are obtained based on the bi-directional nearest neighbor matching method.Finally,initial matches are refined by removing outliers using the RANSAC algorithm.2.A new registration method combining point and line features is proposed.First,the affine transformation parameters of the images are obtained based on point feature matching.Then the stable line features are extracted based on the LSD algorithm.After matching the lines using both lines’ attributes and geometric constraints,the high-precision registration of the images are realized with the edge points on the corresponding lines using the general photogrammetry.3.Applying the phase congruency-based corner screening and feature matching method to the registration of the large-view infrared band images,and the experimental results show that the method can achieve correct registration between bands,and has greater advantages in efficiency and accuracy.Applying the new registration method combining point and line features to the registration between infrared mosaic image and DOM.The experimental results show that the spatial accuracy of infrared mosaic image can be improved to less than 5 pixels.