An edge preserving median filter for images based on level-sets

We propose an edge preserving median filter, called the level-set adaptive median filter, for noise removal in images. This filter uses connected sets of pixels with the same value, namely level-sets, as flexible regions which contour to edges in the image. The filter determines whether a set is noise or signal and smooths the noise. These set regions are flexible in terms of shape since they are created based on their values, and being data-driven therefore provide the mechanism for the filter to preserve edges in the image. We used metrics such as Pratt's Figure of Merit and Peak-Signal-to-Noise Ratio on the labelled faces in the wild data set. We concluded that the proposed level-set adaptive median filter does remove noise while preserving the edges in the image better than the traditional adaptive median filter.