Image registration through self-correcting based on line segments

Image registration system is able to judge whether images are taken from the same scene in aerospace field especially for the aerial work. The traditional related work has some problems in registration accuracy and robustness to image distortion. To tackle these problems, a novel self-correcting scheme for image registration based on line segments is proposed. We use the line feature instead of traditional point feature to determine the accurate correspondence between images and this brings in strong robustness to image distortion. The paper introduces two algorithms containing a rough registration algorithm and an accurate registration algorithm to improve registration accuracy. In order to search optimal registration correspondence in the appropriate candidate lines, a novel optimization algorithm called line segment length histogram method is proposed. Rough registration can obtain an approximate registration according to the candidate lines and it indeed improves the efficiency. In accurate registration algorithm, the line feature is expressed in a different form and can avoid more drawbacks in registration. Then the accurate algorithm can find all the corresponding line segments in images and their numbers can reflect the quality of our registration work. We have tested many images and all of them can match accurately with each other by the computed optimal correspondence. The experimental results demonstrate the high accuracy and strong robustness of our algorithm on various images.