Enhanced Bag-of-Features model for image classification

Bag-of-Features (BOF) representation is a very popular model for content based image classification. In BOF, term frequency (tf) and inverse document frequency (idf) is a very popular model to compute the weights of the visual vocabularies. However, tf-idf model does not contain the class information of images. Fortunately, chi-square model contains the class information well. So, in order to enhance tf-idf model, we adopt chi-square with an exponent to add the image class information to the tf-idf model, as a result we call this enhanced model tf-idf-chi-exp model. We extract scale invariant feature transform (SIFT) features from images and carry out the experiments and then the outcomes show that the tf-idf-chi-exp model performs better than tf-idf model. According to different environment, if we can set a reasonable exponent of tf-idf-chi-exp model its performance will be somewhat better.