4

Does anyone know of a (purely) matlab implementation of Haar feature extraction (the kind used in Viola&Jones object detection algorithm)? (I ran across a matlab binding to Opencv's implementation but this is not what I am looking for)

Amro
  • 121,265
  • 25
  • 232
  • 431
elijah
  • 1,684
  • 2
  • 12
  • 15
  • It might help to know what you need it for. Do you simply want to *use* it in MATLAB, or are you looking for something you can study to better understand how the algorithm is implemented (with MATLAB being the preferred language)? I'm also not sure how to interpret "purely". Does that mean you want only m-files, or that you want to be able to use it without needing anything other than MATLAB (and its toolboxes)? – gnovice Jan 13 '10 at 17:00
  • by using 'purely' I was hoping to avoid such references where the crux of the algorithm is c and the use of matlab is only as invocation (that is what i meant by 'binding'). The purpose is to use the code as a basis for further research. Matlab code would be more convenient for such purposes. – elijah Jan 13 '10 at 17:33

1 Answers1

2

I came across this code published as part of a Google Summer of Code 2007 project under GNU GPLv2 license.

Face detection for F-spot

Note that I haven't tested it myself, just thought it could be useful for you..

Amro
  • 121,265
  • 25
  • 232
  • 431
  • Thanks, at a glimpse it seems promising, i'll dive into it and update. – elijah Jan 14 '10 at 09:20
  • It is an implementation of the general Viola and Jones cascade framework, but unfortunately it seems that the actual features used are not Haar features. – elijah Jan 14 '10 at 10:34
  • well it shouldnt be very complicated to try and implement that part yourself. Maybe this question http://stackoverflow.com/questions/1707620/viola-jones-face-detection-claims-180k-features could help you in generating the haar features – Amro Jan 14 '10 at 10:51
  • you are right it should not be too difficult. I was hoping someone already implemented it nonetheless – elijah Jan 14 '10 at 13:56