Abstract
Previous research has shown that the gist of a visual scene can be understood in as little as 100ms (Potter, 1975; Oliva & Torralba, 2001). Here we used Repetition Blindness (RB) to investigate what underlies this rapid processing. RB is the failure to report the second instance of a repeated item in a rapid serial visual stream (RSVP) of information. It is thought to reflect the repeated activation of a memory representation ("type"), but a failure to individuate the repeated items into distinct visual episodes. We sought to determine what aspects of a stimulus contribute to type activation, by varying the level of similarity (visual feature vs categorical) between critical items. Participants viewed RSVP streams containing either two or three scene images preceded and followed by masks. We manipulated the relationship between the first and last scenes, such that they were: 1) identical repeated images, 2) mirror-reversed versions of the same image, 3) different members of the same category (e.g., two different beaches), or 4) scenes from different categories (non-repeats). Across different experiments, we measured participants' accuracy in reporting the scenes or their sensitivity to detect repetitions, using a range of presentation rates (106ms – 153ms/item). We consistently failed to find RB for scenes. In general, participants were more accurate in reporting repeated scenes compared to non-repeats (i.e., showed repetition advantage), and showed better sensitivity to detect repetitions of the same scene. This was true for both identical and mirror-reversed repeated scenes, though the advantage for mirror-reversed scenes was only apparent with longer exposure durations. In contrast, category repeats were no different from non-repeated scenes. These results suggest that early processing of scenes relies on the visual features present in the image, rather than higher-level categorical information.
Meeting abstract presented at VSS 2014