Abstract
Dynamic objects are effortlessly recognized across motion, viewpoint, and articulation. We investigated the perception and neural representation of dynamic objects using fMRI with both GLM and multi-voxel pattern analysis (MVPA). First, we identified brain regions recruited during dynamic object perception using a novel localizer which contrasted moving objects with phase-scrambled versions of the same moving objects. Static objects and scrambled objects were also included as comparisons to traditional LOC localizers. Results identified regions of occipito-temporal cortex largely overlapping with regions selective for static objects, but subsuming a larger area extending more dorsally to include the hMT+ complex and parietal regions. Second, we conducted several experiments to investigate the coding of dynamic objects over motion and articulation. Subjects viewed multiple example animations of novel articulating objects, which varied across viewpoint, size, and motion path. One experiment included three objects with eighty example animations for each; a second experiment included six objects with forty examples for each. A SVM pattern classifier was trained on one set of example animations, and tested on a different set of examples – examining whether a tested brain region encodes dynamic objects across motion, articulation and viewpoint. Classification was performed in ROIs identified with the dynamic object localizer and a motion localizer that identified hMT+. Results showed above chance classification in all ROIs, and chance classification in a control ROI in frontal cortex. A second assumption-free analysis employed a whole-brain searchlight classifier to search for areas that might encode information about dynamic objects. This searchlight identified additional regions that overlapped with parts of early visual cortex (determined by retinotopy scans), dynamic LOC, and hMT+. Our results indicate: that multiple areas of early and higher-level visual cortex are recruited during dynamic object perception and many of these areas encode information that is invariant across motion and state of articulation.
Meeting abstract presented at VSS 2013