In visual bag of words model, I have been able to construct the visual codebook through kmeans clustering of SIFT descriptors. How to calculate the feature vector for an image then?


For each image, we can find interesting SIFT points, and for each points we have a SIFT descriptor (which is usually a 128 length vector).

im1 ==> SIFT feature f1 (10 by 128) (here 10 is an abitrary number) im2 ==> SIFT feature f2 (20 by 128) ...

If we combine all SIFT features, f=[f1; f2; ..] and perform kmeans clustering we will get the codebook c=[c1; c2; .. c10] which is bow codebooks.

From the codebook how can we find the feature vectore, represent image im1?

  • 1
    $\begingroup$ This is too little information. Can you elaborate? $\endgroup$
    – Phonon
    Nov 29, 2013 at 1:58

1 Answer 1


The feature vector is just the histogram of how many times a feature from each cluster appeared in the image.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.