The Maximum Homogeneity Neighborhood Node lays 8 different areas around one Pixel and applies a measurement of homogeneity which is then processed in some sort of weight. Then, it calculates each areas mean-value and combines both mean-value and weight to the new pixel value, so that the most homogeneous areas take the most part in the new value of the pixel and areas that lay over edges are not taken into account. Via a parameter you can regulate the influence of the weights.