Correcting radial lens distortion using image and point correspondences

This paper describes two new methods for lens distortion calibration using image and point correspondences. Images (or feature points) captured by a camera are undistorted and projected into a calibration pattern image. Both methods apply the Gauss-Newton-Levenberg-Marquardt non-lineax optimization technique to match, in one case, the camera image and the pattern image, and in the other case, selected point correspondences from the camera image to the pattern image. One way to automatically find good point correspondences is presented. Experimental results compare the performance of both methods and show better results using point to point correspondences.