Hugo Turro-Serrano¹ (), Lasyapriya Pidaparthi¹, Frank Tong¹; ¹Vanderbilt University

In everyday vision, many objects will appear partially occluded due to the presence of foreground objects. Amodal completion allows us to perceive partially visible objects by “filling in” the missing information caused by occlusion (Nakayama et al., 1990). Previous studies have investigated perceptual completion of partially occluded objects using 2D displays (Johnson & Olshausen, 2005; Tang et al., 2018), but due to the lack of stereo-depth cues, it remains unclear how these findings would generalize to realistic 3D conditions of visual occlusion. In our study, participants (n=17) viewed stereoscopic grayscale images drawn from eight possible animal categories, which were fragmented by textured horizontal bars. These bar occluders appeared either in front of, behind, or at the same disparity relative to the object. To investigate how perceptual completion unfolds over time, the object images and occluders were briefly presented for either 27, 53, 107 or 213 ms. Subjects pressed a key to identify the object category. We found that participants were more accurate when the stimulus was presented for longer durations (p<.001) and when the occluder appeared in front of the object (p<.001). Importantly, the effect of occluder depth on recognition accuracy increased with stimulus duration (p=.015), requiring 107ms for foreground occlusion to have a significant beneficial effect. Surprisingly, recognition was no more accurate for occluders at zero disparity than at far disparity even though the latter condition was highly unnatural. Our results suggest that object completion based on 3D depth perception requires time to take effect, presumably because of the recurrent or top-down interactions that are required to support amodal completion. Together, our findings add to our knowledge of how the visual system leverages stereoscopic depth cues to complete partially occluded objects and facilitate object recognition in our daily lives.

Acknowledgements: This research was supported by National Eye Institute grant R01EY035157 to FT, P30EY008126 to the Vanderbilt Vision Research Center, and Vanderbilt’s MARC undergraduate program supported by grant T34GM136451 from the National Institutes of Health.