Automated digital image registration and its parallel implementation

Digital image registration is a fundamental task in image processing, which is concerned with establishment of correspondence, particularly geometric correspondence, between two or more pictures taken, for example, at different times, from different sensors or from different viewpoints. Because of the variety of the gray levels in images, it's very difficult to match them automatically with a satisfactory accuracy. In this chapter, we address the problem of geometric registration, in which a spatial transformation is needed to remove the variations of the misaligned images. This geometric registration problem is discussed in a theoretical modal first, and then a novel efficient geometric registration algorithm based on the shape of the closed-regions is presented. Experiments have verified the advantages of this algorithm, but shown that the performance of its sequentially execution depends too much on the size of the input images. Its time complexity increases exponentially as image size increases, so finally we extend the sequential algorithm to a parallel scheme to perform the registration task more efficiently.