The Research Based on Multi-view Image Registration

Abstract

Digital image registration is a fundamental problem in computer vision. It has a wide range of applications in three-dimensional image reconstruction, object recognition, object classification, camera calibration, and other aspects from. Digital image registration is usually a digital image processing pre-processing stage, such as digital image fusion, image integration and so on. Digital image registration techniques can match and stack different images which are got from in the same scene with different conditions to generate a new interpretation about this scene, which can not be taken from single image information. To get more rich and detailed image information, we often get the same scene from different angles to shoot, and through this shooting we get a number of different perspectives of the image, which is Multi-View Images. This paper presents a multi-view image registration method. This method is used by the Harris feature detection feature extraction of multi-perspective image, and then to extract the feature point neighborhood of the characteristics of the same region normalized so that the Multi-View Image Registration in transformation into a local rigid transformation occurred in the image registration problem, and using SIFT descriptor on the characteristics. Finally calculate the similarity distance function to get feature matching points.