Image saliency mapping and ranking using an extensible visual attention model based on MPEG-7 feature descriptors

Abstract

In visual perception, finding regions of interest in a scene is very important in the carrying out visual tasks. Recently there have been a number of works proposing saliency detectors and visual attention models. In this paper, we propose an extensible visual attention framework based on MPEG-7 descriptors. Hotspots in an image are detected from the combined saliency map obtained from multiple feature maps of multi-scales. The saliency concept is then further extended and we propose a saliency index for the ranking of images on their interestingness. Simulations on hotspots detection and automatic image ranking are conducted and statistically tested with a user test. Results show that our method captures more important regions of interest and the automatic ranking positively agrees to user rankings.