Correction of geometric lens distortion through image warping

This paper presents a new technique for estimating and correcting the geometric distortion produced by common off-the-shelf lenses. The task of recovering the three most important parameters modeling the lens distortion of a digital camera is formulated as a problem of image warping where the distorted image of a planar calibration plate taken by the camera has to be registered onto the virtual undistorted image of the same target that would be produced by an ideal pinhole digital camera in a convenient position with respect to it. The Levenberg-Marquardt algorithm is chosen for solving the associated minimization problem in which the cost function is the squared /spl Lscr//sub 2/ norm of the difference between the intensities of the reference and distorted image. In this paper, the algorithm is applied to a practical case of lens distortion correction, the experimental results reported here confirm the effectiveness of the image warping approach.