Causal Saliency Effects During Natural Vision. Carmi, R. & Itti, L. In Proc. ACM Eye Tracking Research and Applications, pages 11-18, Mar, 2006. Recipient of Best Paper Award
abstract   bibtex   
Salient stimuli, such as color or motion contrasts, attract human attention, thus providing a fast heuristic for focusing limited neural resources on behaviorally relevant sensory inputs. Here we address the following questions: What types of saliency attract attention and how do they compare to each other during natural vision? We asked human participants to inspect scene-shuffled video clips, tracked their instantaneous eye-position, and quantified how well a battery of computational saliency models predicted overt attentional selections (saccades). Saliency effects were measured as a function of total viewing time, proximity to abrupt scene transitions (jump cuts), and inter-participant consistency. All saliency models predicted overall attentional selection well above chance, with dynamic models being equally predictive to each other, and up to 3.6 times more predictive than static models. Among static models, color contrast was up to 2.1 more predictive than intensity variance. These results establish the superiority of dynamic over static saliency in attracting attention during natural vision, while also indicating a special role for color. We propose that purely bottom-up or purely top-down saccades are rare in real world environments. Instead, attentional selections are typically determined by dynamic interactions between bottom-up and top-down influences, which are sometimes cooperative and sometimes competitive.
@inproceedings{ Carmi_Itti06etra,
  title = {Causal Saliency Effects During Natural Vision},
  author = {R. Carmi and L. Itti},
  abstract = {Salient stimuli, such as color or motion contrasts, attract
human attention, thus providing a fast heuristic for focusing limited
neural resources on behaviorally relevant sensory inputs. Here we
address the following questions: What types of saliency attract
attention and how do they compare to each other during natural vision?
We asked human participants to inspect scene-shuffled video clips,
tracked their instantaneous eye-position, and quantified how well a
battery of computational saliency models predicted overt attentional
selections (saccades). Saliency effects were measured as a function of
total viewing time, proximity to abrupt scene transitions (jump cuts),
and inter-participant consistency. All saliency models predicted
overall attentional selection well above chance, with dynamic models
being equally predictive to each other, and up to 3.6 times more
predictive than static models. Among static models, color contrast was
up to 2.1 more predictive than intensity variance. These results
establish the superiority of dynamic over static saliency in
attracting attention during natural vision, while also indicating a
special role for color. We propose that purely bottom-up or purely
top-down saccades are rare in real world environments. Instead,
attentional selections are typically determined by dynamic
interactions between bottom-up and top-down influences, which are
sometimes cooperative and sometimes competitive.},
  year = {2006},
  month = {Mar},
  pages = {11-18},
  booktitle = {Proc. ACM Eye Tracking Research and Applications},
  type = {mod;bu;td;eye;psy},
  review = {full/conf},
  file = {http://ilab.usc.edu/publications/doc/Carmi_Itti06etra.pdf},
  note = {Recipient of Best Paper Award}
}
Downloads: 0