I'm confused about how to calculate feature scaling in Viola Jones algorithm. For example in "An Analysis of the Viola-Jones Face Detection Algorithm" of Yi-Qing Wang, he proposed the following for feature type "a":
1. set the original feature support size a ← 2wh
2. i
Germán,
The formula it is a little bit confusing.
The idea is to obtain a w relative to the canonical size of the rectangular filter on the 24x24 pattern, and the size of the Sliding Window (ROI). At the same time, the value of w must correspond to a filter lying inside the ROI.
For your question about the threshold. In fact, the output values of the rectangular filters of different sizes are not exactly the same, even if we apply feature scaling. But, in practice, the difference is not important for the the classification and you are not obliged to change the threshold values.
If you are interested, you can find a Matlab implementation of Adaboost in:
http://pablonegri.free.fr/Downloads/RealAdaboost_PANKit.htm
Regards (Un abrazo),
Pablo
Pablo,
Thank you for your answer. I was confused with the algorithm.
I will see your blog.
Regards,
Germán
- Badges
- Science topic
- Similar topics
- Psychology
- Cognitive Science
- Mind