A morphology-based filter structure for edge-enhancing smoothing

We introduce the value-and-criterion filter structure, a new framework for designing filters based on mathematical morphology. The value-and-criterion filter structure is more flexible than the morphological structure, because it allows linear and nonlinear operations other than just the minimum and maximum to be performed on the data. One particular value-and-criterion filter, the mean of least variance (MLV) filter, finds the mean over the “subwindow” of data with the smallest variance within an overall window. The ability of the MLV filter to smooth noise while preserving and enhancing edges and corners is demonstrated. An example application of the MLV filter in improving the contrast of magnetic resonance images is also shown