Bag of Words Classification Using KNN

Question

I have computed bag of words models for training and testing images. I have 260 bow vectors(100x1) for training images and 282 bow vectors (100x1) for testing images. I would like to classify the test images by using knn algorithm. However, I don't know how to use those bow vectors.

cagatayodabasi cagatayodabasi · Accepted Answer · 2015-05-15T01:18:16

I assume that you want to use KNN in your decision process.

To able to use KNN, you need to calculate a distance between two vectors. You can use the norm to calculate distance. Fortunately, MATLAB is doing this for us if you have Statistics and Machine Learning Toolbox.

Let X be a vector and the each row of it is your 1x100 BOW vectors(transpose of them). and y be a vector that assign the class of each BOW vector. For instance, if you want to classify the images whether they includes bicycle or not, your y must contain binary(if bicycle is presented in image : 1 or Otherwise: 0) information about each histogram.

x = [ - ---- -- - -- -  first histogram;         y = [1;
      - - - ---- -- --  second histogram;             0;
       - ---- - ------  third histogram]              1]

mdl = fitcknn(X,y); %this will be your model.

Actually, I don't know whether it will work or not with BOW, because I always use SVM with it. So, good luck and please inform us if it worked or not.

Bag of Words Classification Using KNN

1 Answers