Real-Time Restoration of Lens Distorted Images by Digital Image Processing

This paper describes a real-time system for image acquisition, where images are corrected for lensdistortion. Three kinds of distortion can be corrected by the system: barrel, pincushion and pocket-handkerchief. Applications can be found in very high light-efficient systems where lenses with a large numerical aperture produce barrelor pincushion distortion. Examples are medical X-ray systems, where pincushion distortion is introduced by the curved output screen of the image intensifier. Another example can be found in digital copiers and scanners. Since distortion does not blur the image, but merely destroys the proportionality between image and object coordinates, it can be corrected by digital image processing. A method for correction is discussed, where a third ordertwo dimensional polynomial is used. An algorithm to determine the parameters for this polynomial will also be addressed. Because the correction requires resampling of the input images, a cubic spline interpolation technique is used to obtain the output image. The system has to process 25 frames of 512x512 pixels, while 16 pixels are used for interpolation of one output pixel. So, high processing power and I/O bandwidth are required. Therefore instead of using DSP’s, we have designed two dedicated VLSI-circuits. Both circuits will be presented in this paper.