Abstract
Humans are highly proficient at interpreting images of other humans—an ability that challenges current models given the wide variation in body appearances and configurations. Previous studies have emphasized the configural processing of whole bodies using simplified stimuli. However, it remains unclear how local and structural information among body parts are integrated. To answer this question, we examined human ability to identify natural images of pairs of body parts presented in static circular apertures, given various amounts of spatial context and structural information between body parts. In a series of online experiments, observers were asked to identify one of six target body parts in the presence of another “context” part. The target parts were fixed in size. In Experiment 1, the informativeness of the context parts, relative to the target parts, was manipulated by varying their sizes and spatial relations (retained or disrupted) or skeletally connected or not. In Experiment 2, the context apertures were rotated at angles from 0 to 180 degrees while maintaining the same size. Experiment 1 found that retained spatial relations facilitated the disambiguation of local uncertainty whereas disrupted spatial relations interfered with target identification. The accuracy increased monotonically with the size of the context parts. In Experiment 2, for skeletally connected pairs, recognition accuracy decreased linearly with rotations from 0 to 90 degrees and stopped dropping after 90 degrees. Whereas for unconnected pairs, there was an approximately linear drop from 0 to 180 degrees. Our study showed that humans efficiently use structural knowledge to resolve local ambiguities in natural images of pairs of parts. Furthermore, human observers are highly sensitive to proper relative spatial relations, not merely to the semantic association of a related nearby part, but also to low- and mid-level cues to alignment and connectedness, indicating strong prior knowledge of body part spatial relationships.