Rotation-tolerant representations elucidate the time-course of high-level object processing

Autor: Denise Moerel, Tijl Grootswagers, Amanda K Robinson, Patrick Engeler, Alex O. Holcombe, Thomas A. Carlson
Rok vydání: 2022
Popis: Humans have little difficulty recognising visual objects in many circumstances, despite the very different retinal images that result from different viewpoints. One source of variability is 2-D rotation, where an object seen from different perspectives results in different orientations. Here, we studied how the brain transforms rotated object images into object representations that are tolerant to rotation. We measured time-varying electroencephalography responses to objects in eight orientations, presented at either 5 Hz or 20 Hz. We used multivariate classification to assess at what point in time rotation-tolerant object information emerged, and whether we could disrupt the rotation-tolerant object processing by presenting stimuli rapidly (20 Hz) to limit the depth of processing. We compared this to fixed-rotation measures of object decoding, where the classifier is trained and tested on the same orientation. Our results showed that both fixed-rotation and rotation-tolerant object decoding emerged at an early stage of processing, less than 100 ms after stimulus onset. However, rotation-tolerant information peaked later than fixed-rotation information, suggesting rotation-tolerant object representations are most robust during a late stage of processing, around 200 ms after stimulus onset. Both fixed-rotation and rotation-tolerant object information was lower for the 20 Hz compared to 5 Hz presentation rate, which suggests that object information processing is disrupted, but not eliminated, for fast presentation rates. Our results show that object information arises at similar times in the brain regardless of whether it is investigated with the fixed-rotation or rotation-tolerant object decoding method, but it is the later stage of processing that reconciles different viewpoints into a single rotation-tolerant representation.
Databáze: OpenAIRE