A Model of Saliency-based Visual Attention for Rapid Scene Analysis-英文文献.pdf

1 A Mo del of Saliencybased Visual Input image Attention for Rapid Scene Analysis Linear filtering colors intensity orientations Laurent Itti Christof Ko ch and Ernst Niebur Center-surround differences and normalization Abstract A visual attention system inspired by the b ehav ior and the neuronal architecture of the early primate visual Feature maps system is presented Multiscale image features are combined (12 maps) (6 maps) (24 maps) into a single top ographical saliency map A dynamical neu Across-scale combinations and normalization ral network then selects attended lo cations in order of decreas Conspicuity maps ing saliency The system breaks down the complex problem of scene understanding by rapidly selecting in a computationally Linear combinations ecient manner conspicuous lo cations to b e analyzed in detail Saliency map Index terms Visual attention scene analysis feature extrac Winner-take-all Inhibition tion target detection visual search of return