MeAdian robust spatial filtering on satellite images

The initial aim with the MeAdian was to obtain a local spatial filtering method which enables to process a robust centre based on a combination of the mean and the median. The MeAdian is an auto-adaptive filter that tends to the mean when this one is more robust, to the Median otherwise. The MeAdian, including or not the covariance of the Mean and the Median, remains one of the most robust estimators faced to different distributions, due to its auto-adaptive capabilities. In this paper, we improve the MeAdian filtering for contour detection in image analysis. The results show the double effect of the MeAdian: a combination of smoothing and planing according to the local distributions encountered. The MeAdian tends to define areas with high homogeneity, like the median does, but whose borders are smoothed or antialiased, like the mean does.