Image cluster with visual aesthetic

Most image cluster algorithms group images based on the content, that can be described with some low-level image features or geometry features. In this paper, we proposed novel grouping criteria for image cluster from a aesthetic perspective. By studying the skills and techniques of professional photographers, we introduce several image features that can describe the styles of photographs. And we quantify these criteria with high-level image features. In particular, Depth of Field(DOF) is computed to describe the aesthetic characteristic of a photograph for the first time. The proposed scheme has advantages in clustering images whose themes are similar or same but photographic styles are different. This often happens when an user organizes web search results with an individual preference. In the last, the experiments are carried out to show the effectiveness of the proposed method.