Global featureless estimation of radial distortions

We consider the problem of image registration in the case of a radial deformation around an unknown center. We propose a novel approach for the estimation of the radial distortion parameters, which is applicable for camera calibration and image correction. The parameters of the deformation consist of the center of distortion (which is not necessarily the center of the image) and the parameters of the distortion function. We propose a closed form explicit solution that performs well for a wide family of radial distortion models and reduces the non linear problem to that of testing a small number of hypotheses. In most distortion estimation methods, some features like points, corners or lines are extracted from the distorted image. In contrast to the feature based methods, our proposed solution uses all of the information in the distorted image which is reasonable considering the fact that the distortion impacts the entire image. Successful experiments of parameter estimation with real and synthetic images with and without noise are presented in this paper.