Classification of a matrix and binary mask

 Hello everyone,
With the aim to optimize the ability to discriminate between two peaks (apes obtained a correlation between a scene containing two letters E and F, and the target that has the letter F) and from a theoretical calculation, I came to this matrix has the form):

M (u, v) = | D (u, v) | cos (Phi (d) - Phi (t))-epsilon | T (u, v) |

with D (u, v) is the FT of the letter E centered;

T (u, v) is the FT of the letter F-centered;

Phi (d) the phase of E centered and Phi (t) is the F center.

* The summation matrix (u, v = 0 to image size / 2)

1-I want to classify this way matrix, where the sum is greater than 0, must I eliminate from the positive area, and if the amount is less than 0, I eliminate from the negative region.

So run the program until you have a sum equal to 0. 2 - How to code binary mask that blocks certain frequencies to optimize the correlation and on what basis, I have to do this?

Thank you in advance