Classification of Material Properties in fMRI E Baumgartner, K R Gegenfurtner |
|---|
Abteilung Allgemeine Psychologie, Justus-Liebig Universität Giessen, Germany
|
| The taxonomy of material categories has previously been investigated with fMRI [Hiramatsu et al, 2011, Neuroimage, 57(2), 482-494]. Here we wanted to explore whether information about material properties can be found in the BOLD response to 84 images showing a large variety of different materials. We asked subjects to rate these images with respect to colorfulness, roughness, texture, hardness, orderliness, & glossiness. We scanned 7 subjects with fMRI while they viewed the images. A linear classifier was applied to visually responsive voxels to discriminate between images with high and low ratings. We found classification accuracy on the fMRI data to be significantly better than chance only for colorfulness (63%, p<0.001), roughness (60%, p<0.05), and texture (64%, p<0.001). Since gloss has received a lot of attention lately, we wanted to look more closely into the representation of glossy materials. We scanned another 6 subjects viewing 58 images of materials selected from a large database of 1492 images as being perceived as very glossy or very matte. A classifier could discriminate the brain activation caused by matte and glossy images with an accuracy of 59% (p<0.01). Our results demonstrate that information about the properties of materials is present in fMRI activation patterns. |
| Up Home |
|---|