Two-Dimensional Global Image Registration Using Local Linear Property of Image Manifold

In this paper, we propose a global image registration method using a sparse dictionary. For global image registration, template images are registered to a pregenerated image stored in a dictionary. Finding the closest image in the dictionary, we obtain a geometric transform between a template image and a reference image. To achieve robustly accurate registration, a large number of transformed images are prepared and stored in the dictionary. To reduce the spacial complexities of this image dictionary, we introduce a method to generate a new template image from a collection of images stored in the image dictionary. This generated template image allows us to achieve accurate image registration, even if the population of image dictionary is relatively small. To the generation of new entry in the dictionary, we use geometrical and topological properties of images in the neighbourhood of a template. This process is achieved using nearest neighbour method. Therefore, to reduce the temporal and spacial complexities of matching pairs in registration, we use a random projection to reduce the size of both template and reference images. Since the random projection approximately preserves topology of a metric space, we compute a matching process in a low-dimensional Euclidean space without significant loss of matching accuracy.