[fprint] Quality of fingerprint matching for a new VFS5011 138a:0011 driver

[Arseniy Lartsev]

Hi,

I've made a driver for
138a:0011 Validity Sensors, Inc. VFS5011 Fingerprint Reader
which I have in my Dell Vostro 3350 laptop. It's now at a stage of "kind of 
working", there are still a few defects left and I'm planning to publish it as 
soon as I fix those. Right now I'm trying to assess the driver's quality.

So I've repeatedly scanned 3 different fingers, 100 times each and calculated 
libfprint's comparison score (as returned by fpi_img_compare_print_data) for 
all pairs of images.

To cut the long story short, for one of the fingers 17% of comparison scores 
between different images of this finger were below 25, 8% below 20, 3% below 
15 and 0.6% below 10. For images of different fingers, 2% scores were ≥10 and 
0.07% ≥15 (the highest was 18, 1 out of 12000).

Do you think it's good enough? Or should I try to improve imaging quality? For 
example, if one enrolls the same finger 4 times (this is what windows software 
on my laptop does, btw) then the chance of false negative would be 0.17⁴ = 
0.08%, and the chance of false positive, as estimated from exponential fit to 
the histogram, would be about 1 in 200000. Does it make sense?

Here are histograms of comparison scores for images of the same finger:
Finger 1 http://mc2-k082.mc2.chalmers.se/fprint/rindex1_cmp.png
Finger 2 http://mc2-k082.mc2.chalmers.se/fprint/rindex2_cmp.png
Finger 3 http://mc2-k082.mc2.chalmers.se/fprint/rmiddle_cmp.png
And here's for images of different fingers:
http://mc2-k082.mc2.chalmers.se/fprint/rindex2_rmiddle_cmp.png

Best regards,
Arseniy