[fprint] Working with small sensors

[Igor Filatov]

> Is this a small sensor problem and is there really too little minutiae?

See https://cedar.buffalo.edu/~srihari/papers/RRP-full.pdf page 7 fig 3 e,f
where it says 20 and 15 minutiae. A 114x57 scan is roughly 1/3 of that
area. In my experience there's usually <5 minutiae on a 144x64 image.
That's way too little. For reliable matching I'd say you want at least 25.

> What are other options to make my device work, maybe by using multiple
images?

At least one driver is doing just that and it's... not the best UX ever
(even when it works) https://github.com/iafilatov/libfprint#common-problems

On Fri, Jun 12, 2020 at 5:05 PM Indev <indev12 at gmail.com> wrote:

> I've heard that fprint matching doesn't work well with small sensors.
>
> In my case (issue #162
> <https://gitlab.freedesktop.org/libfprint/libfprint/-/issues/162>), the
> device returns 114x57 images and there is probably not much minutiae to
> extract. I am certain that it is a press sensor, even though it returns
> multiple images during communication. Despite the fact that it's hard to
> find any specification for the device, Acer Swift 3 promo
> <https://www.youtube.com/watch?v=FgbYrSOy3oQ> shows exactly that.
>
> When running enroll and then verify examples, I get "score 0" on every
> verify.
>
> Basically, my questions are:
>
>    1. Is this a small sensor problem and is there really too little
>    minutiae?
>    2. What is the state of small sensors matching? Are there any working
>    drivers for small sensors?
>    3. How much minutiae is enough to work?
>    4. In my driver code, how can i check minutiae count *before* calling
>    fpi_image_device_image_captured, for example, to prevent submitting images
>    with low minutiae count?
>    5. What are other options to make my device work, maybe by using
>    multiple images?
>
> _______________________________________________
> fprint mailing list
> fprint at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/fprint
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/fprint/attachments/20200613/64a2f425/attachment.htm>