Image Registration Algorithm And The Document Image Registration Applications

Image registration is an important research topic in the fields of computer vision and image processing. It is well known as the precondition of other parts of image processing. A great deal of work is done to study image registration algorithms in this dissertation.For the beginning, two methods of image registration which are the method based on the lightness information and the method based on the features of the images are researched mainly in this dissertation. After the analysis and evaluation of several classic algorithms, a improved method of image registration is summed up.Next, on the base of the above research, a new composite registration algorithm based on image edge feature and transform domain correlation is proposed in this dissertation by combining the algorithm of registration in frequency domain with image feature extraction effectively. The experiment results show that this method can deal with the images existing shift and rotation well.Furthermore, Aiming at the characteristic of document image, a new method of document image registration based on geometric invariance and Hausdorff distance is presented in this dissertation. Experiments indicate our method leads to accurate and robust registration, and can extract the filled-in information reliably.