Determining the receptive field of a neural filter

In this paper, a method for determining the receptive field and the structure of hidden layers of a neural filter (NF) was developed and evaluated. With the proposed method, redundant units are removed from input and hidden layers in an NF based on the influence of removal of units on the error between output and teaching images. By performing the removal of units and retraining for recovery of the loss of the removal repeatedly, the receptive field and a reduced structure of hidden layers are determined. Experiments with NFs were performed for acquiring the function of a known filter, for the reduction of noise in natural images and for the reduction of noise in medical image sequences. By use of the proposed method, redundant units were able to be removed from NFs, while the performance of the NFs was maintained. Experimental results suggested that, with the proposed method, a reasonable receptive field for a given image-processing task could be determined, i.e., the receptive field of the NF trained to obtain the function of a filter corresponded to the kernel of the filter, and the receptive fields of the NFs for noise reduction gathered around the object pixels in the input regions of the NFs.