[fprint] AES1660: call for testers

[Vasily Khoruzhick]

On Fri, Nov 23, 2012 at 11:30 AM, Andreas <krawatten at andreas-loos.com> wrote:
> Hi all,

Hi

> the driver seems to work also for me for practical purposes. Success rate
> for verification is around 50%, which is significantly worse than under
> windows (about 90%).

It's much better for me, success is about 80% for index finger, 70%
for middle, 50% for ring and little fingers :)
Anyway, it's strange, because driver is based on your USB logs, so if
there're some hardcoded values in commands it sends
- it should correspond to your sensor.

> The success rate highly depends on the quality of the model fingerprint in
> the database. So if the first scan in fprint_demo is not good, one cannot
> expect a high success rate (of course!).

Sure, if it has less information to compare - it's harder to consider
if finger is same :)

> Therefore I propose for the future an extension for fprint_demo that makes
> it possible to compose a high quality image from several scans of the same
> finger. Unfortunately, I have at the moment not the time to implement this.

It should be libfprint functionality, and yes, it's quite complex task
(similar to assembling panorama from several photos). Anyway, I'm not good at
image processing algorithms, so some help here would be appreciated.

Meanwhile we could implement following:
do 3 scans during enrollment and choose best one (more minutiaes => good scan)

> Best,
> andreas

Regards
Vasily