Image registration using virtual circles and edge direction

In this paper we combine a novel set of image features called virtual circles with edge direction, to provide an efficient alignment algorithm for image registration under similarity transformations. Virtual circles can be extracted efficiently from many types of images with the help of the distance transform. They can also have extra information, such as their radius and the dominant edge direction in the neighborhood, which can be used for efficient registration. An heuristic called the smoothness criterion is used to choose virtual circles which are less likely to be corrupted by background noise, and to increases the efficiency of the algorithm. Experimental results have shown that this method is highly automatic and has linear complexity in terms of the number of virtual circles extracted.