Any other feature sets beside hu moments to train SVM

asked 2013-08-27 14:31:02 -0500

Vanushka gravatar image

updated 2013-08-27 14:58:50 -0500

berak gravatar image

Are there any other feature sets beside hu moments to train an support vector machine?

Is template matching the best OCR to build a number plate identification system?

edit retag flag offensive close merge delete



hu moments should do well(maybe they're a bit too sparse).

but svm is not at all restricted to that. you could even flatten your images to 1 row, and use that (no, i'm not serious about it)

i.e the code here is using vertical/horizontal histograms of 'on' pixels as features

berak gravatar imageberak ( 2013-08-27 15:06:34 -0500 )edit

Some suggestions of feature sets : average colour values in RGB, YUV, HSV color spaces / Histograms of Oriented Gradients / basic intensity histograms / Gabor filters / Haar - like wavelets / local binary patterns / SIFT SURF ORB FREAK ... 2D features / edge filters / ... The possibilities are endless :)

StevenPuttemans gravatar imageStevenPuttemans ( 2013-08-28 02:11:08 -0500 )edit