Corner Feature Based Research On Registration Between Infrared And Visible Images

Along with the fast growth of the ways and amount of humans'image gaining, some basic researches in digital image processing became more and more important. Among them, image registration is one of the most basic and most difficult fields in computer vision and pattern recognition. Generally, image registration can be defined as to find the same part between two different images as well as the transformation model. Usually the two images are obtained in different times, or under different illumination conditions, or in different resolutions, or from different angles and positions, and the transformation between them could be rigid or affine, etc.Image registration is widely applied in image processing, including image fusion, object tracking, image retrieving, image mosaic, medical image processing, multi-video and multi-sensor image processing. Plus, due to the fast growth of 3D imaging and processing technology in recent years, registration of 3D images is put into use even more frequently. According to the type of images, problems of image registration could be mainly divided into two parts: Mono-model and Multi- model, which have made registration methods and algorithm in a great variety. Through decades'researching in global area, lots of different methods for image registration and matching have been developed. However, there is hardly one or a few solutions that are suitable for the majority, which means different solutions are presented to carter different problems. Therefore, image registration techniques are now far from sophisticated.In this paper, a conclusion about previous achievements in image registration is made. The research of this paper mainly focuses on feature-based algorithms, including a short review of this kind of solutions. Besides, a deep and careful investigation on SIFT (scale invariant feature transform) is done in order to introduce a classic solution model of point feature based algorithms. Meanwhile, a novel feasible solution for registration of infrared image and visible image is presented. Additionally, we provide an instance of real application of image registration technology in image jointing. As feature points and feature descriptors based algorithm are the key idea of this paper, our contribution are mainly:1. After a careful analysis about the popular corner detecting operators, an improvement for Harris operator is made to spread the close corners. More importantly, this improvement could reduce the possibility of miss-match in certain situations;2. A solution that is based on corners and local feature analysis is presented for the fine registration between infrared images and visible images. A novel coarse-to-fine strategy, which is aimed to disperse the miss-match possibility into more than one steps, is the chief idea of this algorithm.3. The combination of multi-features significantly reduced the computing complexity of feature matching and therefore enhanced the real-time capacity.