opencv object detection with my own features

edit

maybe you could show, how your 'fourier descriptors' look like ?

is there any code ?

apart from descriptors, you'll probably need a classification/learning mechanism (like AdaBoost used with haar/lbp cascades), also some 'sliding window' technique or such. what are your plans here (so far..) ?

I am also interested in this

my fourier descriptors are computed globaly in the image, i.e, for each image, one vector descriptor is computed, so for a database of 2500 faces , i have 2500 fourier descriptors ( vector size = 1x4417 ). now, i'm blocked at the training classifier step. thanks

again, please show us, how you calculate those.

then - what exactly is your goal ? detection (face or not) ? classification (who is it) ?

what did you try for classification so far ?

i have already apply this descriptors in a svm context for classification, now my goal is to apply them in object detection context. To compute these descriptors : - i compute the ffteshift of the input image, then i evaluate this fftshift in a regular hexagonal grid points, for each hexgon 6 points, i get a W(f) vector dim 6x2, finally the combination of theses vectors give the invariant descriptor of the image, and it's a vector dim 1xN. (sorry, i can't share the code of computing these descriptors now, it will be after submitting the paper) I tested my descriptors against coil and orl database for classification, and i got accuracy = 97% good classification. Now the next step is to test them in object detection context. Thanks

So your question becomes "How to use Ada Boost cascades on Fourier features?"

Exactly !!

I am afraid you need to modify the cascade code from OpenCV, for that... Or make your own AdaBoost

can you tel me more about how to modifiy the cascade and the training classifier code of opencv, some ideas, ... please

if you want my 2ct, - stay away from the rather outdated and total messy cascade_training code.

rather build your own on top of cv::Boost