[fprint] Aspect ratio of images is wrong?

[Alan Davidson]

I ask because I have a bunch of fingerprints already stored using the
DigitalPersona drivers, and need to be able to compare them to prints taken
with libfprint. I've converted from one data structure to the other, but
they're not matching properly because the distances between corresponding
minutiae differ by so much. The rotating of fingers was to show that the
issue is with libfprint, rather than the DP drivers.

I see a mention on the uru4000 wiki page to Paulo Marques' anti-distortion
code, but the link is dead. A Google search for ["paulo marques"
fingerprint distortion] doesn't turn up anything else obviously relevant.
Is there another copy of that code somewhere?

 - Alan

On Fri, Sep 23, 2016 at 3:37 PM, Martin Hejnfelt <martin at duffman.dk> wrote:

> Hi Alan,
>
> On Fri, 2016-09-23 at 14:23 -0400, Alan Davidson wrote:
> > Hi libfprint folks -
> >
> > I've got a U.are.U 4500 reader. I see in libfprint/drivers/uru4000.c
> > that the image dimensions are hardcoded to 290x384 pixels. However,
> > that doesn't appear to be the right aspect ratio for my hardware.
> > Running fprint_demo, I can see 5 scans side-by-side by enrolling a
> > print. If I rotate my finger 90 degrees between these scans, I see
> > that the horizontal axis is stretched out compared to the vertical
> > one (features that are left-to-right of each other in one image are
> > noticeably further apart than those same features in another image
> > where they're up-and-down). If I enroll my finger vertically and then
> > scan it sideways, the scans don't match.
> >
>
> I think this is what is mentioned on the (rather old) URU4000 drivers (
> https://www.freedesktop.org/wiki/Software/fprint/libfprint/uru4000/)
> page, which also mentions some anti distortion code to deal with this.
>
> However isn't this a rather "thought up" issue, as the unit really
> seems meant to be used in one direction only. I would however think it
> would make a (small?) difference in minutiae detection if the finger is
> not placed straight in the oval direction of the sensor pad.
>
> > I've also been using the closed source drivers from DigitalPersona
> > themselves, which do not have this problem: the distances between
> > corresponding minutiae don't vary as I rotate my finger, and it has
> > no trouble matching fingers that have been rotated. I believe the DP
> > drivers think the resolution of the reader is 354x391 (found by
> > comparing the bytes it records when you save fingerprints to a file
> > with Slide 10 of [1]).
> >
> > Has anyone else seen this? Is there a way to query the hardware to
> > see what resolution images it thinks it records? How can I adjust the
> > libfprint scans so they match fingerprints regardless of orientation?
> >
>
> I am using the unit without issues in a 1:n scenario. I must admit I
> have not tried rotating my finger. Question is if one could find the
> anti-distortion code mentioned (or if it is already there, I have not
> gone through the git history).
>
> Regards,
> Martin
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/fprint/attachments/20160923/14a16676/attachment.html>