On the Design and Study of Adaptive Capability for Neural Network Filters in Image Restoration

Abstract

The design of adaptive filters becomes more important in the field of image restoration. In this paper, we propose a new class of neural network filters which possess more adaptive capability in contrast to that of rank selection (RCRS) filters which are a large class of selection filters. As we know, the RCRS filters possess some powerful filtering capability among the conventional filters for image restoration. The selection filters include the RCRS filters whose output is on of the observation samples. In addition, the RCRS filters need the exponential computation complexity to achieve their design process. Fortunately, the neural network filters can avoid this problem. Finally, the experiment results demonstrate that the adaptive capability for te neural network filters in better than that of the RCRS filters.