Bottom-up saliency model generation using superpixels
Patrik Polatsek, Wanda Benesova
Slovenska Technicka Univ. (Slovakia)
Abstract.Â Prediction of human visual attention is more and more frequentlyÂ applicable in computer graphics, image processing, humancomputerÂ interaction and computer vision. Human attention isÂ influenced by various bottom-up stimuli such as colour, intensityÂ and orientation as well as top-down stimuli related to our memory.Â Saliency models implement bottom-up factors of visual attentionÂ and represent the conspicuousness of a given environment using aÂ saliency map. In general, visual attention processing consists ofÂ identification of individual features and their subsequent combinationÂ to perceive whole objects. Standard hierarchical saliency methodsÂ do not respect the shape of objects and model the saliency asÂ the pixel-by-pixel difference between the centre and its surround.
The aim of our work is to improve the saliency prediction using aÂ superpixel-based approach whose regions should correspond to objectsÂ borders. In this paper we propose a novel saliency methodÂ that combines a hierarchical processing of visual features and aÂ superpixel-based segmentation. The proposed method is comparedÂ with existing saliency models and evaluated on a publicly availableÂ dataset.
Paper will be available in 2015:Â
P. Polatsek and W. Benesova, â€œBottom-up saliency model generation usingÂ superpixels,â€ in Proceedings of the Spring Conference on ComputerÂ Graphics 2015.