Purchase this article with an account.
Laurent Itti, Masatoshi Yoshida, David Berg, Takuro Ikeda, Rikako Kato, Kana Takaura, Tadashi Isa; Role of different salient features in guiding gaze of monkeys with unilateral lesion of primary visual cortex. Journal of Vision 2010;10(7):140. doi: 10.1167/10.7.140.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
In this and the companion paper (Yoshida et al.), we evaluate the role of different low-level visual features in guiding monkey gaze after unilateral ablation of primary visual cortex (area V1). Here we analyze eye movements (108,458 saccades) of six macaques (three normals, three with unilateral V1 ablation) watching 54 minutes of natural videos. A computational model of saliency-based visual attention quantifies the monkeys' propensity to attend to salient targets in their normal vs. lesioned hemifields. The model combines five low-level feature channels: Luminance contrast; L-M (red-green) chromatic contrast; S-(L+M) (blue-yellow) chromatic contrast; oriented edges (V1-like Gabor filters); and motion. All monkeys are significantly attracted towards salient stimuli, with salience computed as the sum of the five features, for saccades both into normal and lesioned hemifields (t-tests, p<0.0001 or better). We conduct an optimization to evaluate the strength by which each feature contributes to gaze, in two steps: (1) find the best set of relative feature weights, for each of three groups: horizontal saccades of control monkeys, horizontal saccades into the affected, and into the intact, hemifields of lesioned monkeys; (2) re-optimize after removing one of the five features (leave-one-out). Comparing (1) with (2) quantifies the non-redundant contribution of each feature to gaze. In all three groups, every feature provides significant non-redundant contribution, with motion the strongest. Significant differences between control and lesioned hemifields are: decreased contribution of Gabor-like oriented edges; increased contribution of luminance contrast; and (surprisingly) increased contribution of S-(L+M) chromatic contrast (Bonferroni-corrected t-tests, p<0.05). The increased contribution of S-(L+M) is further evaluated in the accompanying paper of Yoshida et al. Our results suggest that unilateral ablation of V1 does not abolish the monkeys' natural tendency to gaze towards salient targets during natural vision, but significantly affects the relative strength by which different low-level features contribute to attention.
This PDF is available to Subscribers Only