Abstract
Most previous work on material perception has focused on their optical properties, such as matte and specular reflectance. However, material categories and attributes may also be recognized from their motion flows, such as the contrasting dynamic textures produced by water and honey. We used a multi-arrangement form of multidimensional scaling (MDS; Kriegeskorte and Mur, 2012) to investigate the perceptual dimensions with which human observers compare dynamic textures. Observers used the drag and drop operations of the computer’s mouse to arrange 97 dynamic textures (viewed as animated icons on a computer screen) according to their similarity. The dynamic textures were looped three-second movies windowed with either a small (radius = 38 pixels) or a large (radius = 225 pixels) circular aperture to control the level of spatial context. In the low spatial context condition (small apertures), the subjective reports and the MDS of the similarity arrangements were consistent with categorizations based on color and motion. In the high spatial context condition (large apertures), nine out of eleven human observers clustered the textures into semantic categories, including natural and artificial, liquid viscosity, wind, and hair/pili movement. The MDS of the average similarity arrangements for these nine observers showed that semantic categories were clustered along a primary attribute continuum from highly penetrable (e.g., water, honey), to less penetrable (e.g., hairs, cloth, fire), to impenetrable (e.g., cement, wood, metal). These findings suggest that larger spatial context reduces ambiguity resulting in similarity arrangements based on physically meaningful dimensions.
Meeting abstract presented at VSS 2015