Image registration using the feature-based relational graph

Generally, the geometrical parameters in the conventional image registration system are estimated by applying only one specific image feature, e.g. salient points, contours, or textures, such that the computation complexity may be reduced. However, applying only one specific image feature may be fail for some images because the image characteristics are widely spread among the various kinds of images. Hence, how to select the image feature to estimate the correspondences between two images is an important issue in developing the image registration system. In this paper, a feature-based relational graph for the image features is generated to roughly estimate the correspondence between two images and then the most reliable image feature is determined to calculate the precise geometrical correspondence parameters.