The fact that digital SEM's produce digital images implies that one can use digital filtering to smooth, filter and otherwise manipulate the images. Most of these techniques are convolutional in nature, weighing neighboring pixels with different weights to accomplish a variety of tasks. If only the nearest pixels are used and the weights are all equal, this is simple median averaging. This top image is an unfiltered image of a damaged profilometer stylus. This is a good example as it contains high and low secondary electron contrast and features on different length scales.
The second image is a sharpened image. The convolutional filter is configured to weight nearest neighbors to enhance high spatial frequencies. This enhances or sharpens the fine features in the image. The scratches radiating from the stylus tip on the shank of the stylus are now more clear. It should be noted that this is at the expense of secondary electron contrast on the conical stylus due to enhanced secondary collection on the side of the stylus closest to the ET detector. Despite this reduction of secondary electron contrast, some sense of 3D topography remains.
The final example utilizes image convolution for edge detection. The convolution kernel uses weights of different signs to calculate derivatives from nearest neighbor pixels. The effect of this is to enhance the edges of structures. In this particular case, a Laplacian edge filter was used. Note that most of the secondary electron contrast is lost, as well as all sense of 3D topography, but the scratches and pitting is greatly enhanced.