Matrix Normalization for Haar Decomposition

I've implemented the Wavelet transformation described in the paper below for image retrieval开发者_JAVA技巧.

http://grail.cs.washington.edu/projects/query/

Overall, the system works great but my question doesn't concern the paper at all; rather, the proper way of normalizing the intensity matrix between values of [0, 1].

Normally, I would just loop through the matrix, find the largest value, loop through the matrix again, and then divide by the maximum value. But since I am trying to compare images, shouldn't the divisor be 255 regardless of what the maximum intensity value in the matrix is?

For example, if the maximum intensity value is only 200, when normalizing, do I divide by 200 or 255?

I don't know much about images, but it sounds like you are trying to have two values for two operations: one for comparison across images, and the other normalized for the image itself.

Do you need both? If so, save both. If you only need one or the other, save whichever you need.

Again, not knowing anything about images, I would guess that the cross-image comparison does not require normalized values at all, (unless getting all values between 0 and 1 is a requirement), because the absolute values themselves are fine for comparing one image to another.