Image-based lens geometric distortion correction using minimization of average bicoherence index

摘要

In this paper, a lens geometric distortion correction method for low-cost digital cameras is described. The lens distortion coefficient is estimated by the value minimizing average bicoherence index which denotes image-dependent higher-order correlations in the frequency domain. In order to formulate the minimization of the average bicoherence index in terms of the lens distortion coefficient alone, the feasibility of a square-pixel assumption and the effect of the image center bias on the image quality are shown experimentally. At the same time, a computationally efficient method for calculating the bicoherence index is proposed, which comprises an approximation of coordinate resampling and development of a relationship between the lens distortion coefficients of the original and its reduced image. Depending on the spatial distribution of image features, the performance of the proposed method can be made comparable to that of precision camera calibration.