Image esthetic assessment using both hand-crafting and semantic features

Automatically assessing the visual esthetics of images is of great interest in high-level vision research and has drawn much attention in recent years. Traditional methods heavily depend on the performance of subject region extraction. This paper proposes to use semantic features in the esthetic assessment system because they can implicitly represent the image topic and be helpful if the subject region extraction fails. Accordingly, a framework combining the hand-crafting features with semantic features is proposed to evaluate image esthetic quality. The experimental results show that the semantic features can improve the performance of image esthetic assessment.