Abstract
Theories of visual attention (Treisman and Gelade, 1980; Koch and Ullman, 1985; Itti, Koch, and Niebur, 1998) suggest that the computation of visual saliency, at its early stages, relies on local feature contrasts (e.g. color contrast, orientation contrast) and is carried out separately and independently within each perceptual dimension (e.g., color, orientation, motion, etc.). These theories agree with the general view that different dimensions of visual information are segregated at early stages of visual processing into largely independent visual pathways (Livingstone and Hubel, 1987). This segregation, however, is not clear-cut. For example, many of the cells in the primary visual cortex exhibit similar orientation tuning for both luminance and equiluminant oriented patterns (Johnson and Shapley, 2008). In this study we explore possible interactions across perceptual domains in the computation of orientation saliency. Trials in our experiments were composed of textured backgrounds, whose initial presentation was followed by a brief (150 ms) superposition of the same texture with a single target bar, two target bars, or no bar at all. Bars were oriented parallel or orthogonal to the texture and subjects' 3AFC task was to report the number of bars observed (Mokeichev, Segev, and Ben-Shahar, 2010). In the first experiment, the textured backgrounds and the bars were isoluminant, and a highly significant saliency effects were observed for bars with orthogonal orientation with respect to the background. These results are in agreement with previous reports of isoluminant orientation pop-out effects (Luschow and Nothdurft, 1992). In the second experiment, the textured backgrounds were isoluminant, and the bars were defined by luminance modulation. Here again, highly significant orientation saliency effects were observed. These results point toward interactions between color-defined and luminance defined orientation processing and suggest orientation as a perceptual dimension which is independent of its defining attributes - whether luminance or color.
This research was funded in part by the Israel Science Foundation (ISF) grant No. 1245/08 and the European Union Seventh Framework (Grant FP7-NMP-2009-LARGE-3). We also thank the generous support of the Paul Ivanier center for Robotics Research, the Zlotowski Center for Neuroscience, and the Lynne and William Frankel Center for Computer Science at Ben-Gurion University.