Restoration of geometrically aberrated images using a self-organising neural network

Accurate image positional information is essential in many image processing and computer vision applications such as motion detection, depth-perception and segmentation of different indoor and outdoor scenes. Aberration in images caused due to incorrect lens performance results in serious problems during further processing of these images. The position of different points on the image get shifted during imagery because of imperfect lens design or decentering of lens elements while its manufacture. A new method is proposed in this paper to restore geometrically aberrated images based on Kohonen's model of self-organizing neural network. Different types of aberration including distortion, field curvature and astigmatism can be handled by this method. Simulation of aberrations are incorporated into the a real portrait image for testing purpose and restoration experiments using the neural network model are presented.