Learning Model in Relaxation Algorithm Influenced by Self-Organizing Maps for Image Restoration

This paper presents a learning model in relaxation algorithm influenced by self‐organizing maps for reducing random‐valued impulse noise. Self‐organizing maps have been hitherto studied for the ordering process and the convergence phase of weight vectors. As a novel approach of self‐organizing maps, a learning model in relaxation algorithm for image restoration is proposed, which creates a map containing one unit for each pixel. Utilizing pixel values as input, image inference is conducted by self‐organizing maps. Then, an updating function with threshold according to the difference between input value and inferred value is introduced, so as not to respond to noisy input sensitively. Therefore the inference of an original image proceeds appropriately since any pixel is influenced by neighboring pixels corresponding to the neighboring setting. The characteristic of our approach includes the network constructing the same number of both inputs and weights for inferring the original image from a degraded image in sequential processing, unlike the distinct number in which self‐organizing maps have applied to conventional techniques for the ordering and the convergence of weight vectors. In the inference process, the effect of initial threshold and initial neighborhood on accuracy is examined. Experimental results are presented in order to show that our approach is effective in quality for image restoration. Copyright © 2008 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.