Abstract
Rapid perception of natural scenes has been studied under a wide range of conditions. Yet, little work has dealt specifically with material recognition, so far. Sharan, Rosenholtz & Adelson (VSS, 2009) suggested that the recognition of material categories in real world pictures was remarkably fast and accurate. In contrast, Wolfe & Myers (JOV, 2010) found that visual search for materials was rather inefficient. Therefore, we set out to measure the time course of material categorization in natural images in more detail. We also compared the time course of material categorization with that of object classification. Subjects classified images based on their material or object category. 4 different materials (fabric, wood, stone and metal) were used from the Flickr.com natural image database (Sharan et al., 2009): 50 pictures showing objects made from the material (object condition) and 50 pictures showing close-ups of the material (close-up condition). For the object categorization task 100 pictures were chosen from the COREL database according to the 4 categories: people, means of transport, animals and buildings. Presentation time was varied from 12ms to 118ms. A pattern mask was presented immediately after the image. Subjects were asked to assign each presented picture to one of the given categories in a 4-alternative forced-choice. In line with earlier results, we found that object categorization was extremely fast. At presentation times of 35 ms, observers reached a performance level of 88% correct. The classification of materials was much slower: at 35ms exposure observers performed at 62% correct and they just barely reached 90% correct at 118ms. Similar to Sharan et al. (2009), we found that material recognition was slightly enhanced in the close-up condition compared to the object condition. Overall, object categorization performance was much faster than material categorization performance, indicating that the later might not be that effortless after all.