I am using SIFT features for detecting logos in clothes. SIFT is quite powerful, as it is scale invariant, but how do I deal with the problem of occlusion and warping?

More specifically, the logos in clothes might be partially occluded, and can also be warped when the clothes are stretched or squeezed.

Any suggestion and directions? Thanks in advance!

Some example are showed here:

Logo on leg:

enter image description here

enter image description here

enter image description here

  • $\begingroup$ Are you training a feature classifier with a training data set or are you trying to match to a few exemplars? $\endgroup$
    – D.J.Duff
    Feb 5 '15 at 6:34
  • $\begingroup$ Around 200 examples, and with different types of logos. If training based method works, I can get more examples. $\endgroup$
    – kim
    Feb 5 '15 at 6:42
  • $\begingroup$ You might need to generate lots of images synthetically to get enough data. I don't know, it's not something I've done. I thought your answer might help someone help you. Another possibility is multiscale sliding window like HOG (again you need lots of data), but why don't you try the SIFT and see? $\endgroup$
    – D.J.Duff
    Feb 5 '15 at 20:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.