Automatic image annotation by image fragment matching

Automatic image annotation problem is one of the most important sub-problems of computer vision. It is strongly related to and goes beyond image recognition. The key goal of annotation is to assign a set of words from a given dictionary to a previously unseen image. In this paper, we address two key problems related to image annotation. The first one is low precision of generated answers, the second one is automatic localization of image fragments related to annotations. The proposed method utilizes image fragment matching to precisely localize near-duplicate visual content of images.