Here is my homework 7. In order to find the threshold T for images, I 
use the following intution: Standard deviation of the resulting fourier 
spectrum can give a sharpness measure. Testing this, I saw that the standard 
deviation of the resulting fourier spectrum goes down as the threshold went 
up. Looking at empirical results from my test images, I saw that when the 
standard deviation went down by more than 5% from the original image, the 
image appeared more grainy, but if the standard deviation was within 5%, 
there was no noticeable difference to my perception. The usual result was 
that most of my images could only have a threshold of 10 before a 
significant loss of quality was observed. As my program shows the image the 
fell beneath the standard deviation level, you can see the reduced quality.

    For the second part, by using a high pass filter, the system was run on 
the edges and noise of the picture. Mean was tried, but it proved too 
sensitive to the increasing threshold. As the threshold increases, standard 
deviation may go up at first as some small amount of noise is removed, but 
then it goes down as edge information is removed. As that information is 
removed, the image becomes blurry. The standard deviation on the high pass 
information is a little less sensitive to the increasing threshold than the 
non-high pass information, so the best threshold corresponds to a slightly 
grainier image, but easily recognizable to the human eye.

Hope this is what you had in mind >_>

Thomas Young