DISCRIMINATIVE PARTS DISCOVERY FOR FINE GRAINED OBJECT CATEGORIZATION

Abstract

In the past few years, many object classification techniques have been developed. However, these techniques do not perform very well when categorizing objects belonging to fine-grained categories. It is easier to differentiate between objects belonging to different broader categories as they have different higherlevel features or parts, like differentiating a dog from a car is an easy task. But when it comes to distinguishing between objects belonging to same category, it can be a challenging task as there is very little variance among their parts, like distinguishing between dogs belonging to different breeds. The task is not only to find parts of an object, but to find discriminative parts that help us categorize objects correctly. Recently part-based techniques have shown promising results in fine-grained categorization.