Saliency Detection and Object Classification

Humans have a distinct ability to process only the information that is of interest within a scene, however, this is not an easy task for computers. Trying to replicate this behaviour, many methods have been proposed to generate saliency maps that segment the object of interest within an image. In this paper, we investigate the problem of object classification, and whether saliency detection can be used. We generate saliency maps produced by two different currently published saliency detection methods, and train separate linear SVMs using the feature vectors obtained from these methods. We evaluate these methods against the traditional approach of extracting features from an image for object classification, namely HoG features. Our results show that saliency detection can be used for object classification, and improves accuracy by 5%.