= PDF Reprint,     = BibTeX entry,     = Online Abstract

Click to download BibTeX data Clik to view abstract R. J. Peters, L. Itti, C. Koch, Eye Movements Are Influenced by Short-Range Interactions Among Orientation Channels, In: Proc. Society for Neuroscience Annual Meeting (SFN'02), p. 715.12, Nov 2002. (Cited by 33)

Abstract: Recent research (Parkhurst et al., Vis. Res. 2002) showed that a model of bottom-up visual attention can account in part for the patterns of eye movements made by human observers while free-viewing complex natural and artificial scenes. Using a similar method, we tested an enhanced model with excitatory and inhibitory interactions among units at overlapping locations, tuned to different spatial scales and orientations, as inferred from previous psychophysical experiments involving fine discrimination of gabor-like patches in the periphery (Lee et al., Nat. Neurosci. 1999). Subjects free-viewed images (visual angle 25x20 degrees) from three databases (outdoor photos, fractals, and overhead satellite photos) for 3000ms per image. An infrared eyetracking system (ISCAN, Inc.) recorded eye position at 120Hz with a spatial precision of 0.5 degrees. For each image, we computed the mean model-predicted salience of the points traversed by each subject's scanpath, and judged these values by their z-score in a distribution obtained from random scanpaths of similar length. Across all conditions, the z-scores ranged from 4-14, confirming that in general, our model of bottom-up attention predicts human eye movements with high statistical significance. Moreover, the addition of interactions among oriented units with overlapping receptive fields led to a robust increase in the z-scores, both overall and for individual subjects and image databases. Thus, these interactions, originally modeled after simple gabor-like stimuli viewed under covert attention, also appear to contribute to subjects' overt eye movements under more natural free-viewing conditions.

Themes: Computational Modeling, Human Psychophysics, Model of Bottom-Up Saliency-Based Visual Attention


Copyright © 2000-2007 by the University of Southern California, iLab and Prof. Laurent Itti.
This page generated by bibTOhtml on Tue 09 Jan 2024 12:10:23 PM PST