ADAPTIVE REGULARIZATION IN IMAGE-RESTORATION BY UNSUPERVISED LEARNING

We present a neural network based image restoration approach which mak es use of multiple regularization parameters within the same image to achieve adaptive regularization. We derived an unsupervised learning s cheme that estimates the appropriate parameter value for each pixel si te based on information provided by the dynamics of the neural network . This scheme is based on the principle of adopting small regularizati on parameters for the highly textured regions to emphasize the details , while using large regularization parameters for the smooth regions t o suppress the more visible noise and ringing in those regions. A seco ndary parameter update process is incorporated after the primary image gray values update process, to determine the appropriate local parame ter at each image pixel. The variables required in the adaptation equa tions of the auxiliary neuron arise as byproducts of computational res ults of the primary neuron. As a result, the determination of the regu larization parameter only involves local computation and no explicit e valuation of the underlying cost function is required. The current alg orithm was applied to a number of real world images. The results corre spond closely to our original expectation in that the algorithm automa tically adopts small regularization parameters for the highly textured regions while maintaining high parameters for smooth regions, thus re sulting in an overall pleasing appearance for the restored images. (C) 1998 SPIE and IS&T. [S1017-9909(98)01801-7].