Hierarchical identification of visually salient image regions

The saliency map model proposed by Itti and Koch has been a popular method in explaining the guidance of visual attention using only bottom-up information. The method makes one-level salient-point extraction, and does not take human visual resolution into account. We propose a hierarchical architecture to identify salient regions in a multiple-layer manner. Two ways of attention movements are introduced to mimic the psychological process of human vision: depth search and within-level position shift. A visual attention tree (VAT) is constructed to help guide human visual search that does not take a definite route. The proposed method makes full use of information at different scales and produces satisfactory results in salient region extraction.