Abstract
Although it has been shown that many visual cortical areas process binocular disparity, it is still not clear how these areas are involved in stereoscopic 3D surface perception. In this study, we conducted fMRI experiment to provide evidence whether ROIs are involved in 3D surface by comparing the accuracy of classifying different 3D surfaces using multi-voxel pattern analysis and further investigated what levels of process for the surface perception are involved. The ROIs were a defined by standard retinotopic mapping in early visual areas and localizers for higher areas. Random dot stereograms were used to depict hemi-cylindrical convex and concave surfaces. There were two types of hemi-cylinders: horizontally positioned defined by shear disparity and vertically positioned defined by compression disparity. We also showed each type of surfaces at two different depth positions to investigate whether the process of the surfaces is independent from the depth position. Linear support vector machines were trained to classify whether the stimuli were convex or concave. Two types of accuracy assessments were performed to investigate the generalized representation of 3D surfaces: (1) Same-type stimuli validation: trained on surfaces defined by shear or compression disparity and tested on the surfaces defined by the same type of disparity. (2) Cross validation: trained on surfaces defined by shear or compression disparity and then tested on the surfaces defined by the different type of disparity. Results showed that while classification accuracies were significantly higher than chance level only for same-type stimuli in V1 and V2, V3A showed classification accuracies were significantly higher for both same-type and cross validation. These results suggest that V3A is related to more generalized process of 3D surface perception irrespective of different depth positions and different disparity types, while V1 and V2 are related to low level process of disparity information.
Meeting abstract presented at VSS 2018