Just as was done to create the hot pixel mask, use the 5x5 median blur transformation. This
replaces each original pixel in the image with the pixel that is the median average in the
original's 5x5 neighborhood. If the original pixel was hot, it will be replaced with one that
better characterizes the local statistics of the image.

The behavior of the median blur transform is preferable to using a Gaussian blur. A Gaussian blur
transform dilutes the color of any hot pixels (and therefore appears to accomplish much of what we
need), but does not remove the anomalous color. A single red pixel in a green background will be
replaced with a green pixel by the median transform, whereas a Gaussian blur replaces the red pixel
with one that is mostly green, but retains a small amount of red, depending on the radius of the blur.

The median blur transform is less "damaging" to the original than a Gaussian blur is.

For this image, I run the median blur transform two times, because some of the hot pixels have
relatively large (but faint) halos around them.