>>>Did anyone ever have a look at the software these guys use in real life? I very much doubt that rotating name label is essential to catching a criminal. Or that thousands of image refreshes are actually helping the fingerprint matching app.
>>
>>Exactly my thought. Why display each print as it goes through the comparator? The human eye can't follow, and, like you say, it would slow up the search. But then again, I think we've established that we're at one on this subject :-)
>>
>>Also, they scan in a dab from the scene, and the s/w auto. orientates and resizes it to those on the AFIS. I'm sure a human would be needed to ID, isolate and orientate the section of sample.
>
>If I was writing the app, I'd make it orientation independent. It actually matches the characteristic points and their orientation - not by absolute position, but to a position relative to each other. I'd rather store the angles between any three of them and their relative distances and forget about absolute position and general angle.
>
>But then when the match is found, it's always straight up :). They either have to match its direction when scanning (which I doubt - they gather it on tape at any old angle that's physically available), or that the app which accepts the scan actually normalizes it somehow and puts the scanned image straight.

Exactly. The scanned image (from any old angle) would, I presume, need at least a polar orientation adjoint. If you look at a "partial" of your own dabs, there's no way of knowing whether those whorls and swirls are the top or side or what of the whole print.

>
>It's actually even harder for the facial recognition software - the shot may have been taken at any angle, so it can only match the relative proportions between the facial features, regardless of the angle. It probably also gets a lot of help from the image refreshes :).

Not to mention foreshortening and the 2-D effect of the photo