FRACTIONAL ORDER MEAN IN IMAGE-PROCESSING

We explore the features of a family of nonlinear filters called the fr actional order mean. The median, the mode, and other statistical chara cteristics belong to this family as particular integer orders. The beh avior of these filters in noise suppression in images is investigated. An improved version of these filters, which we call the 8-pixel windo w mean filter is also explored. In this case, the filtered value of th e pixel is compared with its experimental value in the presence of a p reset threshold. Noise suppression performance is tested with the usua l mean square error (MSE) criterion, with a correlation criterion, and with a new criterion, the mean relative error (MRE). Evidence is foun d that the MRE correlates better than the MSE with the perceived visua l quality of the processed image.