The influence of partial occlusion on shape recognition

Humans are able to recognize objects from their outlines, largely irrespective of perspective, position and scale. Previous studies have drawn different conclusions regarding the importance of specific features such as curvature maxima in shape recognition. Schmidtmann et al. (2015) demonstrated that if observers were asked to match a segmented shape, which contained only either convex, concave curvature maxima or intermediate segments for varying lengths, to one of two subsequently presented re-scaled and re-positioned whole-contour shapes, they performed significantly better for convexities compared to the other shape features. Surprisingly, performance for convexities was independent of the length of the segments. These results implied that closed planar shapes are encoded using the positions of convexities, rather than concavities or intermediate regions, and that a simple shape-template model could explain performance (Schmidtmann et al., 2015, Scientific Reports, 5, 17142.). 

In natural viewing conditions, however, objects are often partially occluded. To investigate the effect of partial occlusion, I employed the same paradigm and stimuli as previously (Schmidtmann et al. 2015), but occluded either 16.7%, 33% or 50% of the shape (i.e., only half the shape was visible in the 50% condition). 

Results show that performance for convex features is superior to the other shape features and independent of segment length, replicating previous data. More importantly, however, recognition performance is only significantly impaired when 50% of the shape is occluded. These results demonstrate the importance of shape convexities for shape encoding, and the flexibility of the visual system to deal with partially occluded shapes.

https://ava2020meeting.wordpress.com

Schmidtmann, G., The influence of partial occlusion on shape recognition, AVA X-MAS Meeting 2020 [SLIDES]

Posted
AuthorGunnar Schmidtmann