I imagine that the date and time character images would have very different fft2 characteristics than the rest of the image.
Depending what you can count on to be in the date image, you might be able to do some simple detection of vertical or horizontal lines to get yourself in the right area.
If you do an erosion operation you would get rid of the specs in the image. bwlabel the graph after that, and look at the regionstats for each of the areas. You should be able to rule out some of the areas from the statistics.
If you were to dilate after the erosion, then you might perhaps join the entire date and time into a blob (or two blobs) that would likely be larger than anything else. Once you know the part of the image that contains the characters, it should be easier to remove the text.
Mind you, I still like ImageAnalyst's answer in the newsgroup, namely that the date and time are probably in pretty much the same place in each image, so find the area once manually and then either zero out that entire box or focus your removal efforts on just that part of the image.
