Abstract
Object recognition involves different steps of visual processing, such as detection, categorization and identification. A previous study on object recognition reported counter-intuitive results that basic-level categorization occurred as accurately and fast as object detection (Grill-Spector & Kanwisher, 2005). However, we often experience that object detection is faster than categorization. We suspected that the null results found in the previous study might be due to either lack of sensitivity in their measurements or insufficient modulation of task difficulty. In Experiment 1, the exposure duration was more finely varied from 1 to 5 frames (1 frame: 11.7ms) as compared to the previous study. Object detection performance was better than basic level categorization when the exposure duration was from 2 to 4 frames. In Experiment 2, we measured thresholds for each task. To measure thresholds, percentage of visual signal was manipulated by scrambling images. The threshold was defined as the level of visual signal in which participants detected/categorized images. Participants' threshold (38.8%) in the detection task was significantly lower than that (48.5%) in the categorization task. In Experiment 3, we increased the task difficulty by introducing positional uncertainty. Specifically, the location of the target to be detected/categorized was randomly chosen among multiple locations. We simultaneously presented 100% scrambled images in the rest of the locations. The number of locations was either 2, 4, or 8. Participants' performance decreased as the number of locations increased in both tasks. More importantly, overall accuracy in the detection task (95.3%) was significantly higher than that (67.9%) in the categorization task. Across all three experiments, we found that object detection performance was better than the basic-level categorization. These results suggest that object detection occurs before categorization when either the sensitivity of measurements or the level of difficulty increases.
This work was supported by the Korea Science and Engineering Foundation (KOSEF) grant funded by the Korea government (MOST) (No.M10644020001-06N4402-00110).