The goal of this research is to develop automatic and efficient algorithms for image registration under similarity transformations, which include translation, scale and rotation. Automatic registration means that manual intervention by the user is minimized as much as possible, while efficient means having the lowest complexity possible and practical execution times.; The main contribution of this work is a novel set of image features called virtual circles and their use in the registration of images under similarity transformations. A virtual circle is a circle with maximal radius encompassing a background area that does not contain edge points. The motivation behind this idea, is the use the empty spaces between the edge points, as features, instead of the edge points themselves. However, since these regions can not be easily described, circles are used to model them. Since these circles do not exist in the original edge map, they are called virtual circles.; Virtual circles have many nice properties such as the radius, the direction attribute, and so on, which can be used for efficient registration. Furthermore, virtual circles are frequent and can be extracted efficiently with the help of the distance transform from many types of images. One drawback of virtual circles is their vulnerability to local variations. However, through the use of a heuristic called the smoothness criterion, virtual circles which are less likely to be corrupted can be selected.; We have tested the new virtual circles method in the registration of 66 pairs of images of printed labels, circuit boards, and various indoor scenes. Experimental result have shown that this method has practical execution times for relatively large transformation ranges. It is also highly automatic, because it has a small number of parameters, which almost never had to be changed throughout the experiments. In addition, it is flexible and can terminate in many ways. For example, it can terminate when available running time is up, when all possible pairings are tested, or when a the target similarity measure is reached.
展开▼
