CATEGORIZATION OF SIMILAR OBJECTS USING BAG OF VISUAL WORDS AND SUPPORT VECTOR MACHINES

This paper studies the problem of visual subcategorization of objects within a larger category. Such categorization seems more challenging than categorization of objects from visually distinctive categories, previously presented in the literature. The proposed methodology is based on "Bag of Visual Words" using Scale-Invariant Feature Transform (SIFT) descriptors and Support Vector Machines (SVM). We present the results of the experimental session, both for categorization of visually similar and visually distinctive objects. In addition, we attempt to empirically identify the most effective visual dictionary size and the feature vector normalization scheme.