There are several things to point out in analyzing our data.

First, the correlation values vary from bill to bill. These values actually mean comparatively little, as long as you get the right answer. The solution, as far as the program is concerned, is the one with the highest overall correlation (as long as it is above 0.8500). That said, there are a couple of things to be noted about the back of the $5 bill. Its correlation was the lowest in the group, at 0.8918. The original scan of this bill was very noisy with some very interesting colors. Even with the noise and distortion, it still found the right answer.

That's hardly the case for the front of the $5 bill, which represents our only failure. First, it is important to note the similarities between the front of the $5 bill and the $1 bill. They are almost the same thing, and actually have a close correlation to each other. The reason that this bill failed is due to there being an edge around the outside, that was the result of it being poorly cropped when initially scanned. This extra edge confused the corner finding algorithm, which misplaced the corners. Oddly, the bill correlated more to a $1. We then took the $5 image, cropped off the offending corner, and performed another test. This time, we received the correct 0.9109 correlation that you see in the data sheet. For experimental value, we compared the corrected version of the $5 back with the $1, just to see how close they really are. The result is not surprising: 0.9082. In other words, the difference between the $1 and the $5 is about 0.0027.