Fast And Accuracy Registration Of Multispectral Image

With the continuous development of multispectral sensor imaging technology in recent years, multispectral images are playing a more and more important role in the military, aerospace, medical and other fields.Images with the same scene obtained under different spectrums contain complementary information. Making full use of the complementary information requires the accurate image fusion technology. Image registration is the key technology in the process of high quality of image fusion, and a fundamental problem in computer vision and image processing. It is the process of aligning two images with the same scene acquired by different sensors, different time, different perspective through a certain space transformation relationship to the same coordinates.Currently, the main image registration method is to detect keypoint from the two images, and then construct a local feature descriptor which describes the neighborhood information of the keypoint. The keypoint based method performs more robustness, and has achieved good results in the single-spectral image registration. However, due to the large difference in the imaging mechanism of multispectral, even the same scene shows different under different spectra. The performance of local feature in multispectral image registration is significantly reduced. In view of this, our paper aims to improve the accuracy of registration on the basis of existing technology. However, the improvement of registration accuracy will inevitably bring about the increase of computational complexity, and how to reduce the time required for registration is also an important problem.EOH descriptor is designed to solve the problem of multispectral image registration. However, it couldn't register the images with rotation transform well. In this paper, we define a main orientation for each descriptor, and make it have some adaptability to the rotation of the images.In order to construct a descriptor with only local features in multispectral images, we use the edge points in the image to construct the global features in the process of matching the feature points and measure the quality of the matching pairs. However, the introduction of global features brings large computational complexity. In order to improve the speed of the algorithm, this paper proposes a coarse to fine matching method, which divides the registration process into multiple steps, and selects the matching set with higher accuracy as the evaluation standard.Experiment results show that the construction of global features can effectively improve the accuracy of image registration, and the coarse-to-fine registration method proposed in this paper also reduces the computational complexity brought by the global feature and improves the registration to a certain extent of the accuracy.