I was reading these slides about Bag of Features (BoF). At slide 23 you can read:

each image is represented by a vector, typically 1000-4000 dimension, normalization with L1/L2 norm

Why we should normalize the feature histogram vector for image-classification/retrieval applications?

In addition, the distance used for normalize histograms should be the same for computing the distance between them? Because the standard distance used for BoF histograms is $\chi2$ distance, not L1/L2.


Since you did not get any answer yet, let me give you a quick one.

Normalizing vectors leads to transforming them to unit vectors. Unit vector is a vector in normed vector space that has length of 1.

Saying it in plain English, you can calculate length of a vector using vector norm, and by dividing vectors by their lengths you make them have the same length. This makes comparison between vectors and computation easier. If you want analogy, it is like dividing the counts with total count and receiving percentages that always range from 0 to 100% no matter of the counts. You can check the linked Wikipedia entries and if still you find all the vector stuff too complicated you can check Khan academy lectures on linear algebra for some introduction.


Your Answer

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

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