Restoring images with a multiscale neural network based technique

This paper describes a neural network based multiscale image restoration approach using multilayer perceptron neural networks trained with artificially degraded images of gray level co-centered circles. The main goal of the approach is to make the neural network learn inherent space relations of the degraded pixels in restoring the image. In the conducted experiment, the degradation is simulated by filtering the image with a low pass Gaussian filter and adding noise to the pixels at preestablished rates. Degraded image pixels make the input and nondegraded image pixels make the target output for the supervised learning process. The neural network performs an inverse operation by recovering a quasi-non-degraded image in terms of least squared. The main difference of the present approach to existing ones relies on the fact the space relations are taken from different scales, thus providing correlated space data to the neural network. The approach attempts to develop a simple method that provide good restored versions of degraded images, without the need of a priori knowledge or estimation of the possible image degradation causes. The multiscale operation is simulated by considering different window sizes around a pixel. In the generalization phase the neural network is exposed to indoor, outdoor, and satellite degraded images following the same steps used to degrade the artificial image of circles. The neural network restoration results show the proposed approach is promising and may be used in restoration processes with the advantage it does not need a priori knowledge of the degradation causes.