search

iafilatov / libfprint

libfrpint driver for a family of Elantech fingerprint sensors
https://cgit.freedesktop.org/libfprint/libfprint/
GNU Lesser General Public License v2.1
146 stars 20 forks source link

Unrecognized error code when running img_capture with Elan 04f3:0903 #21

Closed neslinesli93 closed 5 years ago

neslinesli93 commented 5 years ago

Hi, thanks for the project - awesome stuff!

I have been trying to setup the whole libfprint/fprintd stuff today on my zenbook ux430. I've tried to run your examples as well, and while verify and enroll seem to work perfectly, I had a strange issue when running img_capture.

I am pasting the logs here:

Found device claimed by ElanTech Fingerprint Sensor driver
Opened device. It's now time to scan your finger.

sync:error [fp_dev_img_capture] unrecognised return code 101
image capture failed, code -22

It looks like fp_dev_img_capture is getting this weird 101 error code when calling fp_async_capture_start... maybe it's a driver issue?

Thanks again!

iafilatov commented 5 years ago

Hi,

Could you please post the entire output of img_capture?

neslinesli93 commented 5 years ago

That is the entire output... should I try to compile with the debug flag to see if there are more logs?

iafilatov commented 5 years ago

Oh I see.. please try elan-glib-2.50 branch, or better yet, master from https://gitlab.freedesktop.org/libfprint/libfprint

neslinesli93 commented 5 years ago

Okay, I used the libfprint master, and this is the log img_capture produced when I quickly put my finger on the sensor and then remove it: https://pastebin.com/dTbpKgcv The interesting line is 225

iafilatov commented 5 years ago

You need to swipe. The driver aborts because there's not enough frames (line 204).

neslinesli93 commented 5 years ago

Why is swiping needed at all? Please be patient, I'm just trying to understand a bit more what libfprint is about and how it's implemented. I use my fingerprint sensor on Windows as well, and swiping is really not needed.

I've skimmed through the code, and it seems that image process stuff is done inside the nbis folder. Maybe that code needs high definition images/more images to work properly, thus the need of having to swipe on the sensor instead of just touching it, so that it captures more data.

After googling a bit about the matter I found this very promising project and its C bindings which, albeit dated, seem more up-to-date than the NIST stuff contained inside the nbis folder.

Do you think it would be feasible to replace the image processing logic of libfprint with sourceAFIS? Using that, we could save multiple templates of the same finger, and use sourceAFIS for fast search among them, hopefully providing way less false negatives/positives when scanning fingerprints.

iafilatov commented 5 years ago

There's a note in readme that describes the problem. It's not about resolution or number of images per se, it's just about the "area" of the print that's available for processing. The whole point of swiping is to expand that area and capture more minutiae.

The algo you link to is also based on minutiae, so I assume it will have the same problems with small cut-out prints as NBIS (which btw will updated to a recent version by libfrpint maintainer in the next release).

Last time I researched this topic there were no open source implementation of fingerprint algos suitable for small sensors like the ones in smartphones and recent laptops. Under Windows/Android/etc., I believe, an altogether different approach is used. It could be based on something other that minutiae or capture some additional properties. If someone actually implemented something like that under a suitable license, it would be a godsend. There's gonna be more and more tiny touch readers in the future and libfprint risks falling behind the curve. When I started this project, I only had a usb reader I bought on ali express. I had no idea they were putting very similar Elan's in ASUS and Xiaomi and that many people would actually need this driver...

If you wish to discuss replacements for NBIS or have other ideas to share, please consider starting a thread on the mailing list so that more people can benefit from the discussion (although you wouldn't be the first one to do so... and we still don't have a solution).