A 2-D filtering structure with neural networks for Gaussian noise cancellation and edge preservation in images

In this paper, we propose the structure of a two-dimensional adaptive digital filter for cancellation of white Gaussian noise in images, and edge preservation. This structure is composed of three parts. We use a two-dimensional filtering algorithm to avoid disturbance due to one-dimensional filtering, a neural network to update the filter coefficients, and a variable-size filtering window to preserve edges. Experimental results show that a filter with the proposed two-dimensional structure cancels the white Gaussian noise and preserves the edges of the image better than those filters based on a one-dimensional filtering algorithm or one that does not consider the edges.