Psychophysical study of visual saliency of different hues

Abstract

Visual attention refers to the cognitive mechanism that allows us to select and process only the relevant information arriving at our eyes. Visual saliency models, trying to simulate visual attention and corresponding gaze patterns, have been continuously developed over the last years. (Borji & Itti, 2013). Color information was shown to play an important role in visual attention, and is used in visual saliency computations. Vazquez, et al. (2010) have demonstrated that color features can have a higher contribution to visual attention than intensity features by using gray stimuli composed of random distribution of pixels, where the central region of a stimulus had the same mean, but a different standard deviation varied along several directions in CIELAB. These findings point toward re-assessing the role of color in visual attention, and, consequently, a modified way of using color information in visual saliency models. In our experiment observers were asked to view stimuli on a calibrated monitor and report a number of detected color patches presented at random locations on a masking gray background of the same lightness. The patches were produced as random distribution of pixels with target means and standard deviations for hue angles, lightness and chroma. The hue angles were constant within each stimulus. Observers' eye movements were recorded via an SMI remote eye tracker and used for the analysis in conjunction with the reported data. The red patches (hue angles 10° and 30°) had significantly lower reported chroma values compared to green patches (130° and 150°), while being significantly less frequently fixated than the green patches. A similar discrepancy existed for blue (250° - 270°) and yellow hues (90°): the threshold reported chroma values were lower for blue than for yellow hues, while significantly more fixations occurred on yellow patches. The results can be interpreted in support of a dissociation between attention and consciousness as proposed by Koch and Tsuchiya (2007)>.

Meeting abstract presented at the 2016 OSA Fall Vision Meeting

Footnotes

 We wish to acknowledge Jeff Pelz for lending the SMI eyetracker, Reynold Bailey for sharing a Matlab code, and Mark Fairchild for fruitful discussions.