Abstract
Human face recognition is disproportionately impacted by image rotation, becoming very difficult when an image is viewed upside down. The reasons for this remain a topic of heated debate. One factor might be that variations in three-dimensional (3d) structure are very important for facial coding. Inverting a facial image would therefore result in critical 3d shape from shading cues being viewed from an unfamiliar angle. Here we test this proposition via another manipulation, which eliminates 3d shape from shading cues
altogether. We contrast peoples’ ability to categorize cars and faces when images vary in luminance and when images only contain differences in colour (isoluminance, which eliminates all shape from shading cues). We find that isoluminance impairs performance disproportionately on a facial classification task relative to car classifications. This was true even though the two tasks were behaviourally matched for difficulty when images contained luminance differences. Nor could this effect be due to facial coding being
selectively impacted by image blur at isoluminance, as performance dropped off equally for cars and faces as blur was added to images containing luminance differences. Our data therefore demonstrate that 3d shape from shading cues can be more important for facial coding than for analyses of other object categories.
Meeting abstract presented at VSS 2012