Classification of Synthetic and Real Images Using Pattern Features

[Korean Version PDF] [English Version PDF]

Abstract

Automatic classification of motion pictures has many application areas and one prominent case is image search based on user query. In this particular application, the user’s intention for query is difficult to identify at the semantic level because image search algorithms generally exploit only features of images. In order to address this issue, we propose, in this paper, an automatic image classification algorithm applicable to synthetic and real motion pictures so that a user’s intention can be fully reflected in the image search and classification. Feature-based or edge-based histogram descriptors produce poor results when applied to classification of synthetic and real images because this is inter-class classification. The algorithm proposed in this paper obtains feature vectors of color distribution pattern in motion pictures to classify synthetic and real images. In our experiment, the proposed algorithm was able to classify images with the accuracy standing at around 74%.