[Digikam-devel] face recognition in digikam & sponsoring of face recognition

[Michael G. Hansen]

kunal ghosh wrote:
> hi, i saw lots of feature requests for face recognition in digikam. I am
> working on face-recognition as a project at my college & would like to
> implement the
> same in digikam. Has anybody already started work on it ?

Hi Kunal, hi Juergen,

you both wrote an e-mail about face-recognition, so I think it is best
to summarize my answers in one e-mail ;-)

We had a discussion about face recognition at the coding sprint in
November. The general opinion of the developers who had some experience
with the related algorithms was that the eigenfaces-algorithm used for
example in opencv would probably not be good enough for face recognition
in real world photos, but nobody knew a better approach. Therefore,
nobody started working on face recognition yet. We did however think
about how to annotate regions of images, as that would have to be done
at the same time anyway, but nothing has been coded for that, either.
There is a document outlining the current state here:

http://websvn.kde.org/trunk/extragear/graphics/digikam/project/ImageAnnotation.odt?view=log

I do not have any experience with face recognition algorithms, but since
you (Kunal) are working with them, it might be a good idea to test
whether you can find an algorithm that works well with a set of real
world photos and compare that with other projects out there (Picasa, for
example). You might want to take a look at what was done during a
similar project for f-spot, but I do not know where that code is. Once
we have an algorithm that actually works for real-world photos under
different lighting conditions, one can start integrating it into digikam.

Another option would be to only build a face detection system, to make
it easier for the user to find images with people, and then tag them.
But that would probably not be as cool as automatic tagging ;-)

About the sponsoring of face recognition:
Basically the lack of knowledge about a suitable algorithm is the
show-stopper here. Having a successful fund-raiser only to find out
later that all usable algorithms are patented or proprietary and can not
be used by us would be a bad publicity stunt :-(

About mentoring and related issues, you will have to ask other members
of the team.

Michael