[fprint] AES3500 patch test report

Juvenn Woo machese at gmail.com
Sat May 4 19:21:48 PDT 2013

Yes, Paul, it's my pleasure to do so, and thank you for testing and
evaluating it.

You're right it's a little too small for a press-typed sensor, which could
capture only a sub-area of fingerprint, yet the device works pretty good on
Windows. So I'm inclined to think there could have rooms for improvements.
And, I was not aware of the 90 degree rotation, I think I must have missed
some params to tweak.

Maybe we could later work it out al-together.

I will clean the code a bit against current HEAD of libfprint before I
submit the patch, and will share the docs and toolkits I had used along the

Will let you know then!

I've tested Juvenn's patch as linked in recent message thread on the
mailing list: https://gist.github.com/juvenn/939298 . It bitrotted a
bit wrt to current HEAD, but was trivial to fix.

I should say that I didn't have much first-hand experience with
fingerprint scanning devices. Except for random cases which almost
everyone experienced, like you go to some office, where they have safe
with fingerprint lock, which they can't open for half an hour and
then call a technician with a hammer to work around the technology.

Neither I tried to look up that NIST "NBIS" thingy. But when I was a
young student I registered for access to a handwriting recognition
system from them, which arrived in snail mail on a CD with a nice
printed manual which said that the system achieves recognition rate of
36.578% and its sole purpose is to serve as a baseline for evaluating
other handwriting recognition systems. So, my wild guess is that NBIS
is just the same.

So, I'm not at all surprised with the initial results I got, which
match those posted by another subscriber: I had troubles enrolling a
finger, and then recognizing, while fingerprint images shown by
fprint_demo were pretty legible.

Some thinking helped though. First of all, AES3500 cannot capture a
*fingerprint*, it just physically too small for that. It can capture
only small sub-area of it, then depending on how exactly you put your
finger on it, areas will be rather different, likely straining NBIS.
As a fingerprinting layman, I also put my index finger's tip on
the sensor, and the tip contains almost parallel lines, and probably
lacks enough features to recognize. Shape of my APC Biopod kinda
suggests that they aim for thumb's fingerpad instead, which has curves
and stuff.

All in all, the best results I expectedly had with small finger - the
sensor can capture pretty large part of it. Except that NBIS couldn't
tell the difference between my left and right little fingers.

All in all, specifically AES3500 image capture driver works pretty well
- fingerprint lines are black, spaces between them are white,
grayscale shades are there. The only thing I noticed is that scanned
image is apparently rotated 90 degrees (i.e. mostly horizontal
fingerprint lines are shown vertically), not sure if that presents
additional puzzle for NBIS.

So, +1 for merging the patch, Juvenn, would you submit it via fprint
bugtracker, as was suggested by Vasily?

 Paul                          mailto:pmiscml at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/fprint/attachments/20130505/d51f73e1/attachment.html>

More information about the fprint mailing list