Purchase this article with an account.
Frédéric Poirier, Frédéric Gosselin, Martin Arguin; Equisalience this!. Journal of Vision 2007;7(9):437. doi: 10.1167/7.9.437.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Introduction. Visual saliency plays an important role in early vision, guiding both attention shifts and eye movements. Visual saliency thus forms a central role in many models of early visual processing (e.g. Itti, 2006, VisCog; Zhaoping & Snowden, 2006, VisCog; Wolfe & Horowitz, 2004, NatRevNeurosci). Using a novel psychophysical method to measure saliency, we derive perceptual fields of contextual modulation.
Methods. The stimulus is a grid of right- or left-oblique red or green lines on a black background. Line luminance varies continuously over the image, which participants (N=9) adjust locally towards equisalience using a mouse. Assuming that systematic deviations from equiluminance are indicative of compensation for saliency, local luminance setting correlates negatively with local saliency.
Results. Perceptually salient image regions are more heterogeneous in color and orientation, indicative of short-range iso-feature inhibition. Perceptual fields of context modulation are obtained by correlating image properties with local saliency. Specifically, certain combinations of features correlated with local saliency, the strength of which was dependent on the distance between items containing these features. Using this analysis, we show that: (1) color center-surround fields for different-color are stronger but operate over shorter ranges than for same-color, (2) parallel orientations are inhibited, but less so if continuous, and (3) orthogonal orientations are more salient when end-stopping another line rather than being end-stopped. On average, these perceptive fields predict luminance and account for 60% of the variance in the data.
Discussion. These new results can be compared to predictions from current models of visual salience. Moreover, this new method is sensitive within the normal functioning range, where most current research methods produce ceiling effects and flat reaction time functions. Here, we used a simple stimulus to validate the method, but the method can be generalized to any stimulus (e.g. reading, visual textures, natural images).
This PDF is available to Subscribers Only