IMAGE-RESTORATION USING A MULTILAYER PERCEPTRON WITH A MULTILEVEL SIGMOIDAL FUNCTION

The problem of restoring a blurred and noisy image having many gray levels, without any knowledge of the blurring function and the statistics of the additive noise, is considered. A multilevel sigmoidal function is used as the node nonlinearity, due to which the same number of nodes as in the case of a binary image is sufficient for an image with multiple gray levels. Restoration is achieved by exploiting the generalization capabilities of the multilayer perceptron network. For realistic images, training time becomes a major burden. To overcome this, a segmentation scheme is suggested. Simulation results are also provided.