Fast correction of lens distortion for image applications

Abstract

The main use of video cameras in industrial applications regards to detection of defects and noncontact measuring. If the first case is usually a problem of mismatching between two images, the absolute position being nonimportant, the second relies on accurate proportional correspondence between image and object dimensions. In the usual set-up, the object is mainly flat and its image is obtained using a video camera whose lens axis is orthogonal and centred relative to the object. The co-ordinate system of the object should be ideally related to that of the image by a simple translation and change of scale, once an initial mechanical rotation adjustment of the set-up is made. The geometric distortions, mainly pincushion and barrel types, must be either avoided, using a well corrected lens in a proper set-up, or corrected through digital image processing techniques. The use of an acceptable lens imposes a higher price or restrictions to the angle of view, as wide angles are harder to correct acceptably. When space limitations are important and a wide angle is sought, an expensive lens and/or time consuming correcting algorithms are mandatory. The proposed algorithm builds a correction table for the set-up in use, which remains fairly constant in industrial applications. This table divides the image in regions. For each region the correction can be put as a single delta-x/delta-y translation of the included pixels. This can be done fairly quickly when compared to the general morphing algorithms.