OPTIMIZATION OF THE WEIGHTED MEDIAN FILTER BY LEARNING

The weighted median filter (WMF) is a generalization of the median filter. The WMF is more effective for image processing than the conventional median filter. However, the design of the parameters of the WMF is a difficult problem. A novel method of optimizing the WMF is proposed that utilizes the close relation between the nonrecursive WMF and the feed-forward neural network with shift-invariant weight coefficients. The optimization problem of the WMF results in the learning of the interconnection weights of the network.