Abstract
A lively Dunedin street scene, and a panoramic view of the Southern Alps - two images that might appeal and interest a viewer. But where do people look, and which of those images appears more interesting? In this paper, we are introducing a visual attention model based on MPEG-7 descriptors that creates multi-scale feature maps to detect interest hotspots in images. Further, we are assessing three methods that use attention models for image ranking and compare them to results gathered in a user test. Preliminary results indicate that rankings created by our model show a high agreement with rankings obtained in a pilot user study.