Study On Automatic Registration Of Visual/IR Images

The registration of visual/IR images is one of the key technologys in terminal guidance of new-type cruise missile. Based on summarizing the existing registration algorithms and analyzing the characters of the visual/IR images, the paper focuses on the problems of visual/IR images.The first two chapters of this paper expatiate on the background and significance of the topic and summarize the existing image registration algorithms. Through analyzing the imaging characteristics of the visual and IR sensors, the paper maily discusses the registration algorithms based on feature.In order to solve the problems in the previous registration algorithms based on feature, chapter III presents a global optimal algorithm to register visual/IR images. This algorithm consists of five steps: 1) to extract line segments; 2) to pair the line segments into Main-Line-Pairs (MLPs); 3) to find out all MLPs that may correspond with each other; 4) to find the optimal registered subset using branch-and-bound method; 5) to compute model parameters using the vertexs of the corresponding MLPs as control points and complete the registration of the images. The algorithm has two main contributions: 1) a new registration feature is devised that is named MLP. It can be used to get the corresponding points easily and has aboundant information and can be easily extracted; 2) it proposes a new strategy for iamge registration, which takes into account image similarity and space compatibility.Chapter IV proposes an image registration algorithm, which combines the control of the model with Gray-level Information. The algorithm consists of four steps: 1) to extract point features; 2) to register point features using the control of the model and Gray-level Information; 3) to compute model parameters using registered point features; 4) to resample the sensed image according to the gained model and complete image registration. The method uses the prior model in the process of finding corresponding points, so that low compute complexity and high registration precision are acquired.Chapter V proposes a semi-automatic high resolution image registration algorithm, which has three main characteristics. Firstly, the algorithm is undependent with the transform model and only requires that the transform holds the global compatibility. Secondly, in the process of registration only a few control points are chosed manually to compute the initial parameters, which will be computed accurately basing on anomaticly extracted points. For that, the algorithm has low compute load and fast speed and high precision. At last the algorithm is adapted with the registration of various images from same or different sensors.Chapter VI designs and realizes an image registration system which includes the algorithms proposed in this paper, and uses lots of real images to test the algorithms. The results prove that the algorithms could match visual/IR images effectively.