There is another method of low-pass filtering that is even better at eliminating noise. Instead of replacing a pixel with the average of its neighbors, it is replaced with the median of its neighbors. A median is calculated by sorting the values in increasing order. The value in the middle of the list is the median. Half the pixels are brighter than the median, and half are dimmer.
With a median filter, any pixel that is much different from its neighbors is eliminated. It therefore suppresses ”impulsive” noise, such as hot pixels and cosmic ray hits, very strongly. Unlike the other filters described so far, the median filter is non-linear. This has some disadvantages; for example, an attempt to produce an unsharp mask with a median filter can result in artifacts – structures that do not exist in the actual image.
A better approach for outlier pixels is to repair just those pixels. There are several approaches. One is to use the Kernel Filter hot/dead pixel removal filters, which detect and replace bad pixels with the average of the surrounding pixels. They are very effective and only affect pixels that exceed the threshold, so their overall effect is much better than a median filter. A similar effect is achieved using the Remove Bad Pixels command, which does the replacement using a map of ”known bad” pixels.
The best method to eliminate hot pixels does not use a filter per se, but does use a Median calculation. You take a series of exposures, and ”dither” the camera position slightly between each exposure (MaxIm DL can automatically dither using the autoguider, telescope control, or AO-7 control). You then combine the images using Median, Sigma Clip, or SD Mask combine. This very effectively suppresses hot pixels without artificially substituting pixels.
The Rotational Gradient filter is intended for use on objects which are mostly circularly symmetrical, to extract detail from non-symmetric features. A typical application is highlighting jets from the core of a galaxy or a comet nucleus. By changing the angle of the filter, you can emphasize details with differing angular size.
Local Adaptive Filters selectively emphasize image features that are lower contrast, by boosting brightness differences in areas that have low standard deviations compared to the average brightness level. The Local Adaptive Filter is very effective for extracting faint details from planetary images, but it can cause excessive noise amplification if over-applied.
Rank Filters take the area around a pixel, sort the values within that area, and then pick out a value according to its percentile (ranking) in the list. These filters strongly emphasize small differences between adjacent pixels.