0

I am attempting product classification based on image processing. I will have (say 10000) images of, say 5, different types of object and my system will segregate those. As a solution, I am trying to extract select haar features and the location on the image where these features were detected. I will then train a classifier(to be decided) on these fields.

I am concerned about the time it will take to extract all haar feature at every location in the image (unlike face detection wherein absence of one feature eliminates the need to find other features in that window). I have my naive function which detects feature by usual way of summing in integral images but it is suboptimal. Is there a way I can do this computation faster (some library etc). Language/Tool is not a barrier. Thanks.

Anurag Priyadarshi
  • 1,113
  • 9
  • 17
  • 1
    If the features are discriminative in this case, there's no reason why an early reject cascade shouldn't work as in face detection. If they are not, the runtime is unimportant. – Photon Jun 25 '15 at 20:42
  • Yes, as far as classification is concerned, I will use reject cascade.But my initial task, before classification, is finding which features are important for a product class. So I need to find all features and then train the classifier to identify important features. Please let me know if I need to elaborate. – Anurag Priyadarshi Jun 26 '15 at 08:50

0 Answers0