A Method Of Image Registration Based On Linear Features And Mutual Information

Image registration is a matching process of two or more image that is taken from different time and different sensors or different perspectives of the same scene. It is widely used in computer vision, pattern recognition, remote sensing image processing, medical image processing and many other fields.It is the first problem to solve in image fusion, especially the image data fusion technology Many scholars dedicated to the study of image registration,and has made a series of outstanding achievements now.But there are still many difficult issues, pending further study. The article basically did the following work:First, it introduced the background and significance of image regist-ration and studied the principle and common registration method.Second, it proposed a registration algorithm based on the minimum area bounding rectangle.It rotated the reference image and floating images with equal interval to find the rotation angle when reached the smallest minimum bounding rectangle area; Then according to the minimum area of the bounding rectangle of the two images, we can determine the proportion of scaling for floating image;then it solved the translation problems by the coincidence of image center and figure centroid; finally, seeking the minimum of the absolute value of the two image can give a more accurate translation distance.Third, a combination of line features and mutual information image registration algorithm was proposed. It used MSLD line describes transforing the three-dimensional optimization problem into one-dimensional searching along the control line, thus it could speed up the efficiency and performance of the algorithm that based on mutual information.This method used both the advantages of characteristics-based and mutual-information-based algorithm,thus it improved the speed of seeking MI maximum and avoids the local extremum.Fourth,the paper carried on the experiments for the two proposed algorithms above,and analyed the results.Analysis shown that the minimum area bounding rectangle-based algorithm is faster, but the precision is not high enough, and it need to study. The line features and the mutual information-based on registration algorithm is effective,faste and of good robustness.