Feature-based Image Registration And Its Application

Image registration,is a process to match two or more images of the same scene taken at different times,from different view point,or by different sensors,which is widely required in remote sensing,in medicine,in 3D reconstruction,and so on .At the same time, image registration techniques is one of the most difficult and important tasks in computer vision.So far,image registration method is usually divided into three categories: the Pixel-based registration method,the feature-based registration method,the model-based approach.Although the digital image registration technology has many algorithms,but so far not yet an algorithm which can be applied to all images.We have to depend on the specific circumstances.For different image, we should try to find a way which is balanced in speed,accuracy and robustness in pressing,developed an algorithm which is suitable for the image.In this papere,we mainy pay attention to the problem is the accuracy of feature extraction and the robustness of the points matching phase when the image exists a little bit rotation.And put forward a Method For Image Registration Based on The Condition Number And Zernike Moments.In the application part,the main concern is the image fusion method based on the wavelet transforms and improves it .the modified algorithm not only has a good effect in the general image, in the multi-focus image has better results.The main work of this paper is as follows:1.Explained the definition of image registration, theories, methods, processes, analysis the key steps of the point feature matching are to the influence in the results of the registration,summarizes the current major methods of the image registration. Introduced the method of image feature extraction ,and introduceds the common methods of the feature extraction,analyze their strengths and weaknesses. Focuses on the theory of Harris corner detection,and on this basis,introduction the conditions number based on the round-neighborhood to remove some false corners,improving the accuracy and efficiency of the feature matching stage.2. Introduced the status of point matching algorithm and some commonly used local feature descriptors.Focuses on Zernike moments,we introduction the improved Zernike moments as the descriptor of the feature matching.Make the rotated image is still good in the accuracy and robustness. Finally,determine the transformation modol with RANSAC and comliete the registration work3. In the application of section,introduced the basic theory of image fusion,and describes the advantages and disadvantages of various methods in fusion.Focuses on the wavelet transform fusion,proposed the fusion method of the combination of spatial and wavelet transform.Finally,we do an experiment respectively with the image of registrated in this paper and multi-focus image.Finally,this work is summarized this paper,and look to the future in the field of image registration.