Abstract
People can recognize natural objects and natural scenes with remarkable speed, even when they have never seen the pictures before (Biederman et al., 1974; Potter, 1975, 1976; Thorpe et al., 1996; Greene & Oliva, 2008). But how quickly can people recognize natural materials? We built an image database containing 1000 images of 9 material categories (e.g., paper, fabric, glass, etc). To prevent subjects from simply doing object recognition, we used cropped images in which overall object shape was not a useful cue. To prevent subjects from simply using color, texture, or other low level cues, we chose images with highly diverse appearances. For example, “plastic” includes close-ups of red trash bags, transparent CD cases, and multi-colored toys. Images were obtained from websites like flickr.com. We found that humans can correctly categorize images with very short durations and in challenging conditions (e.g., 40 msec followed by a noise mask, or presented in the middle of an RSVP stream at 40msec per image). When we degraded the images by simple manipulations like removing color, or blurring, or inverting contrast, performance was reduced but was still surprisingly good. We also measured recognition speed with reaction time. To measure baseline RT, we gave subjects very simple visual tasks (e.g., Is this disc red or blue? Is this diagonal line tilted left or right?). We then asked them to make a 3-way material category judgment (e.g., paper or plastic or fabric?). Material categorization was nearly as fast as baseline. Beyond judgments of material category, observers can judge dimensions of material appearance like matte/glossy, opaque/translucent, rigid/non-rigid, soft/rough, warm/cool reliably even in 40 ms presentations. In conclusion, material perception is fast and flexible, and can have the same rapidity as object recognition and scene perception. NTT Communication Science Laboratories, Japan National Science Foundation.