Blur detection using a neural network

Image restoration is an ill-posed inversion problem wherein an estimate of the ideal original image is to be extracted from a noisy and blurred observation. The ability to restore such a degraded digital image usually requires accurate knowledge of the blur function as well as additional information on the original image. Unfortunately, such a priori knowledge is not always accessible. This paper describes an iterative scheme for the identification of the blurring by making use of the neural network paradigm and the assumption of physical constraints on the blurring process.