Research On Image Registration Algorithm And Its Applications

Image registration is one of the fundamental tasks within image processing, it finds wile application in medicine, military, remote sensing, astrophysics, geo-physics, and computer vision. For example, image fusion is an important and useful technique in many disparate fields such as image understanding, computer vision and medical image processing, but before the image fusion, all of the images must been registered.The task of image registration is to find an optimal geometric transforma-tion that can transform the images of the same scene but in different coordinate systems, for the reason that they are sampled from different sensors, taken at dif-ferent times or from different viewpoints, into the same coordinate system. Image registration is a problem often encountered in many application. So, in the last few decades, thousands of paper were published on the topic of image registration, but there are still many problems needed to be solved and much works needed to be done. For example, in the methods of subpixel image registration, how to over-come the shortage root in discrete sampling of image also entails a lot of work; other problems such as how to improve the accuracy of image registration, the impact of the interpolating function on the the accuracy of image registration, how to evaluate the result of image registration, etc.This thesis aims at improving the techniques used in image registration, and giving out our novel approaches of image registration. Main contributions of this thesis are as follows.1. Similarity measure is crux of the robust and accuracy of the image registra-tion. The mutual information (MI) is regarded as an efficient similarity measure in the multi-modal image registration. Based on the mutual information and lissajous figures, a novel similarity measure is proposed. In some cases, this simi-larity measure can reduce the area where other similarity measure fail and greatly improve the robust of image registration.2. To improve the robust and accuracy of the image registration algorithm, we firstly proposed a trajectory disturbance phenomenon. With this strategy, the recognition capability of the similarity measure and the accuracy of the registration algorithm based on this similarity measure is greatly improved.3. A novel subpixel image registration approach based on trajectory distur-bance phenomenon and phase correlation is proposed. It can greatly improve the accuracy of the image registration.4. In this paper, a novel image registration method which combines the SIFT(or SURF) algorithm and the contour information is proposed. In some cases, this method is more robust and reliable than the registration methods based only on SIFT(or SURF) algorithm. Also, in this method, an effective criterion to identify the correctness of feature matching is given out.5. At last, we also explore the application of the proposed similarity measure in the remote sensing image registration.The similarity measure based on trajectory disturbance and mutual informa-tion and approaches of image registration propsed in this thesis are essential in both theory aspect and practical aspect. In the future, they will be widely ap-plied to the fields of science research and engineering technology such as military, medicine, remote sensing, etc.