Automatic Image Registration Techniques

Image registration is a fundamental problem in image processing, which is a process to match two or more images of the same scene taken at different times, from different viewpoints, or by different sensors. It is a basic step for image mosaic, object identification, image fusion, detection of changing object, analysis of sequential images etc. Application of image registration is involved in military affairs, remote sensing, medicine and computer vision etc. Through the research of many years, image registration technique had obtained the certain research result. Currently image registration technique can be divided into the handicraft technique and automatic technique. Through the whole process of image registration, there are need no person's help and finished the registration automatically. Automatic technique is the final development goal of the image registration technique. This paper mainly studied the automatic image registration technique. This text mainly completed the following work:When we studied the method of registration based on feature point, firstly, we analyses concept, theory and common methods of registration. And present a method based on SUSAN corner point achieve arithmetic, then study the registration method based on corner points. This paper firstly analyze the theory and method of SUSAN corner point achieve arithmetic, and aim at the bad quality of real time and accuracy, the paper introduce a fast corner point achieve arithmetic. Based on this arithmetic the paper bring forward an improved SUSAN corner point achieve arithmetic. The arithmetic mainly use the gray quality of the pixels around the corner point finish the corner point achieve, i.e. through control the sum and the connectedness of the points which alike the center point finish the corner point achieve. At the same time, aiming at the problem of the threshold T fastness at the SUSAN corner point achieve arithmetic and fast corner point achieve arithmetic, this paper bring forward an automatic method which can settle this problem. Improved SUSAN corner point achieve arithmetic not only settled the real time problem but also settled the wrong corner points problem which in the fast corner point achieve arithmetic. Then the paper use the achieved corner points study the image registration. Firstly study the match of the corner points, and then using this corner point as feature point finished image registration. The experiments show that this method could match tow images perfect. And has higher registration precision.