Infrared Image Mosaicing Research Based On Feature Points

Being an important branch of digital image processing, image stitching technology is in a leading position in various fields. Its main task is to use more than one picture with certain overlap part to form a large, high resolution and seamless complete picture. With the development of science and technology, image stitching technology has been widely used in panorama stitching fields, medical imaging, video retrieval and compression, object three-dimensional reconstruction, and so forth. Especially in the military field, infrared image stitching technology are generally applied into military strategy and tactics such as searching weak infrared target tracking, infrared early warning, target identification and detection and surrounding scan.Firstly, this paper introduces the purpose and significance of the image stitching technology, and its current applications and research. In the meantime, application background of infrared image stitching is also discussed. Then, the basic theoretical knowledge of image stitching, which includes three coordinate systems and the conversion relation of the computer vision and the image of the model transformation, is comprehensively described. In the later part of this paper, the process of image stitching and key technologies used in the stitching process is summarized, and in the meantime, problems that should be noted in the infrared image stitching are proposed.Concerning the infrared image stitching based on feature points, this paper carries out a meticulous research on detection and extraction methods of traditional feature point, and analyzes the advantages and disadvantages of the various methods. And then it analyzes common image registration methods, and with the characteristics of the infrared image, proposes an improved image registration method which combines the phase correlation method with Harris. Firstly it adopts phase correlation method to obtain the amount of displacement, then takes it as a priori information apply to partial Harris corner detector, which will not only shorten the corner detection time, but also reduce the false alarm rate. It utilizes NCC method for feature points matching, then purifies the matching points through higher robustness RANSAC, and then selects the mapping model to calculate the transformation matrix, to achieve the completion of the image registration. Finally, based on the gradual fade-out weighting method, it proposes the improved fusion algorithm in order to achieve seamless requirements. The experimental results show that this method can accurately stitch the rotating infrared image, and improve the accuracy of the original algorithm. Meanwhile, the algorithm has strong real-time feature which is very competitive in the motion compensation of infrared search and track system.