On Tue, Dec 1, 2009 at 10:21 PM, Daniel Drake <[email protected]> wrote:

> Robert Vanak wrote:
>
>> Further research turned out "Fingerprint recognition SDK" which seems much
>> better than Bozorth3, judging from its documentation. I am reconsidering my
>> plans in view of tougher competition. It might be a better idea to improve
>> on top of existing FR-SDK sources, but then that's not so much fun. I have
>> compiled a list of other algorithms with quick evaluation of each here:
>>
>
> Never heard of that one before.


Okay, I investigated Fingerprint recognition SDK in more detail and here's
what I have found:

* much newer than Bozorth3, it's just 2 years old
* sound algorithms that are a considerable improvement over Bozorth3
* more accurate than Bozorth3, uses non-minutia information
* faster than Bozorth3, should be practical to search databases of 10,000
fingerprints
* tunable if provided with database of training fingerprints, just give it
fingerprints from fprint-supported sensors
* tuning can be also used to specialize the algorithm for specific sensors,
especially the small ones
* high tolerance to elastic deformation of fingerprints, this might be
useful for the elderly

I haven't tested it though, but I guess the tests conducted by authors can
be trusted.

See more information about their algorithm here:
http://fingerprintreco.cvs.sourceforge.net/viewvc/fingerprintreco/fingerprintreco/Documents/Presentations/

So my recommendation is to integrate Fingerprint recognition SDK in fprint
as a replacement for Bozorth3.

Regarding my own project, I do not expect it to provide any revolutionary
improvement over FR-SDK. I might do it for the fun anyway, I haven't yet
decided about this.

Robert
_______________________________________________
fprint mailing list
[email protected]
http://lists.reactivated.net/mailman/listinfo/fprint

Reply via email to