Evaluation of an Attractive Region Extraction Method

Abstract

A viewer usually does not perceive everything in a picture. What objects in the picture the viewer perceives depends on the structures in the view, the meanings of the objects, the mental condition of the viewer (demanding, interested, etc.) and so on. If it is assumed that the impressions of viewer are determined by what objects he or she pays attention to, highly attractive regions of a picture would have a lot of information that contribute to the impressions. Consequently, by analyzing those regions, it is possible to predict the impressions of viewer within a reasonable error. This paper proposes a method for extracting attractive regions in a picture. This method extracts attractive regions by evaluating the attractiveness of each segmented region. For the evaluation, we defined an evaluation function based on hue, satulation, lightness and size of the region, as well as, color difference between the region and its surroundings. In our experiments, the proposed method could extract attractive regions precisely.