Bio-Inspired Models of Network, Information, and Computing Systems. 5th International ICST Conference, BIONETICS 2010, Boston, USA, December 1-3, 2010, Revised Selected Papers

Research Article

Contribution of Spatio-temporal Intensity Variation to Bottom-Up Saliency

Download
455 downloads
  • @INPROCEEDINGS{10.1007/978-3-642-32615-8_44,
        author={Eleonora Vig and Michael Dorr and Erhardt Barth},
        title={Contribution of Spatio-temporal Intensity Variation to Bottom-Up Saliency},
        proceedings={Bio-Inspired Models of Network, Information, and Computing Systems. 5th International ICST Conference, BIONETICS 2010, Boston, USA, December 1-3, 2010, Revised Selected Papers},
        proceedings_a={BIONETICS},
        year={2012},
        month={10},
        keywords={video saliency eye movements intrinsic dimension structure tensor natural dynamic scenes},
        doi={10.1007/978-3-642-32615-8_44}
    }
    
  • Eleonora Vig
    Michael Dorr
    Erhardt Barth
    Year: 2012
    Contribution of Spatio-temporal Intensity Variation to Bottom-Up Saliency
    BIONETICS
    Springer
    DOI: 10.1007/978-3-642-32615-8_44
Eleonora Vig1,*, Michael Dorr2,*, Erhardt Barth1,*
  • 1: University of Lübeck
  • 2: Harvard Medical School
*Contact email: vig@inb.uni-luebeck.de, michael.dorr@schepens.harvard.edu, barth@inb.uni-luebeck.de

Abstract

We investigate the contribution of local spatio-temporal variation of image intensity to saliency. To measure different types of variation, we use the geometrical invariants of the structure tensor. With a video represented in spatial axes and and temporal axis , the -dimensional structure tensor can be evaluated for different combinations of axes (2D and 3D) and also for the (degenerate) case of only one axis. The resulting features are evaluated on several spatio-temporal scales in terms of how well they can predict eye movements on complex videos. We find that a 3D structure tensor is optimal: the most predictive regions of a movie are those where intensity changes along all spatial and temporal directions. Among two-dimensional variations, the axis pair , which is sensitive to horizontal translation, outperforms and by a large margin, and is even superior in prediction to two baseline models of bottom-up saliency.