Hi
@Alex : Ok, I tested the eigenface performance a lot... there are a few bugs
that can be ironed out, for example updating the eigenface database with the
exact same face image as before causes a segfault. Also, currently it is
very slow. I can get working on that soon, but I think we can do this
faster.
Have a look at my bitbucket repo "facestalker" :
http://bitbucket.org/aditya_bhatt/facestalker/
In the funcs.h file ( very dirtily coded, sorry ;) ) , I have used some
built-in openCV functions like cvCalcEigenObjects and cvEigenDecomposite
that will calculate eigenvalues, eigenvectors, projections and more.
Using these functions, my facestalker implementation is insanely fast... it
takes about ~5 seconds to train a set of 400 images of size 92x112.
So, what do you say? Should I incorporate some of that code into libface?
--
Aditya Bhatt
My Blog : http://adityabhatt.wordpress.com
|
