Abstract
Previous studies have proposed image based measures of clutter and correlated them to subjective judgments of perceptual clutter (Yu et al., 2014) or threshold contrasts during a search task (Rosenholtz et al., 2007). Here we evaluate multiple clutter metrics (Feature Congestion, FC; Subband Entropy, SE, Rosenholtz et al., 2005; Freeman & Simoncelli, 1995; ProtoObject Segmentation, PS, Yu et al., 2014) and correlate them with the time required for observers to fixate a searched target. In addition, we evaluate the influence of clutter on detectability as a function of retinal eccentricity of the target. Methods: In experiment 1, twelve observers searched with eye movements (free search) for a target (any person; 50 % probability present) in video stimuli rendered from a virtual environment, presented at 200, 400, 800, 1800, or 3200 ms. In experiment 2, twelve observers maintained their gaze on a fixation (forced fixation) with the target appearing at 4 different cued retinal eccentricities (1, 4, 9, and 15 deg) and with varying presentation times (half of free search). For both experiments, observers reported with a 1-10 rating of their confidence about the presence of the target. Results: Times for target foveation in experiment 1 were consistently longer for high clutter scenes irrespective of the metrics (320 ms high clutter vs. 260 ms low clutter), but PS resulted in the highest correlation with clutter (rFC = 0.336 ; rSE = 0.261; rPS = 0.565 for 3200 ms). In experiment 2, a half split analysis (high vs. low clutter) shows clutter degrades detectability only beyond 9 deg. target eccentricity. Finally, the PS clutter metric also resulted in the highest negative correlation with hit rate (r = 0.733). Conclusions: Our results show the effects of clutter on eye movement guidance and target detectability, and suggest that the ProtoObject Segmentation metric can best predict such detriments.
Meeting abstract presented at VSS 2016