Restoring images with a multiscale neural network based technique

This paper describes a neural network based multiscale image restoration approach in which multilayer perceptrons are trained with artificial images of degraded gray level cocentered circles. The main objective of this approach is to make the neural network learn inherent space relations of the degraded pixels in the restoration of the image. In the conducted experiment, the degradation is simulated by submitting the image to a low pass Gaussian filter and the addition of noise to the pixels at pre-established rates. The degraded image pixels make the input and the non-degraded image pixels make the 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 approach to existing ones relies on the fact that the space relations are taken from different scales, thus providing relational space data to the neural network. The approach is an attempt to develop a simple method that may lead to a good restored version of the image, without the need of a priori knowledge of the possible degradation cause. Considering different window sizes around a pixel simulates the multiscale operation. In the generalization phase the neural network is exposed to indoor, outdoor, and satellite degraded images following the same steps use for the artificial circle image. The neural network restoration results show the proposed approach performs similarly to existing methods with the advantage it does not require a priori knowledge of the degradation causes.