An improved algorithm for image registration using robust feature extraction

In this paper a new image registration algorithm for restoring the original size, orientation, and position of a distorted image is proposed. The algorithm is based on three main steps: extraction of some feature points, obtaining the correspondence between the features points of the reference and distorted images, and estimating the transformation parameters that map the distorted image to the reference one. The proposed algorithm uses an efficient feature points extractor which is based on scale-interaction of Mexican-hat wavelets. The correspondence between feature points of the two images is evaluated using the singular value decomposition (SVD) of circular neighbourhoods centered on feature points. The transformation parameters, transforming the distorted image into the reference one, are obtained as the solution of a least-squares minimization problem using the two sets of feature points. Experimental results illustrate the accuracy of registration and the robustness against several common image-processing operations