Blind radial distortion compensation in a single image using fast Hough transform

Abstract

In this paper, we present an automatic technique for compensation of radial distortion, which is characteristic of wide-angle lenses. The proposed method estimates distortion parameters using a single image from unknown source. No calibration objects are required, but it is assumed that the original scene contains straight lines. The method is based on finding such radial distortion parameters, that maximize total length of linear segments. We employ a fast Hough transform to estimate the overall curvature of lines without selecting any. The proposed algorithm is tested on real images obtained using calibrated camera lenses with different radial distortion. For the formal evaluation of the algorithm, we propose a quality measure for geometric distortion compensation, which works correctly even in the case when the problem of determining the coefficients is ill-conditioned.