Image Restoration Using Neural Networks

Abstract

For resolving a restoration problem of degraded and noisy image, we investigate the Hopfield neural network, and we employ the eliminating highest error EHE criterion in intention to improving perfor- mances of network. Moreover, with the purpose to make a better restoration, we take in consideration a human perception in restoration process. To do this, we introduce an adaptive regularization scheme, with contribution of a local statistical analysis, to assigning each pixel one regular- ization parameter regarding to its spatial activity. Due to various values of regularization parameter, this scheme permit us expanding the one network to a network of network NON, which we subsequently elucidate its analogy with the human visual system, a cortex.