Abstract
Are changes that alter the meaning of a visual scene or include the appearance of irregular items more noticeable than other types of changes? Previous work has reported that changes that are “inconsistent” with a scene may be easier to detect (Hollingworth & Henderson, 2003); therefore, we hypothesized that if a change alters the meaning of an image to a greater extent, it would be more easily detected. Likewise, we hypothesized that image changes with greater differences in scene congruence would be more easily detected. To test these hypotheses, we first conducted a large online photo-rating task where subjects answered questions about image pairs where one item was changed in each pair. From these results, we selected image pairs that were defined by one of four features: images with mostly semantic changes, images with mostly incongruent changes, images with minimal semantic or incongruent content (change-present controls), and images that did not change (change-absent controls). Next, in two laboratories, participants completed a change-detection task where on each trial, they viewed images for up to 20 seconds and pressed a spacebar as soon as they experienced a change, which was triggered by a saccade during the viewing. After each trial, they were asked questions about their confidence in their response, their subjective experience of the change, and on miss trials, whether the changed item contained the pre- or post-change attribute(s). Our results showed that detection hit rates positively correlate with some image-feature variables (scene meaning, scene congruence), and negatively correlate with others (scene complexity). Awareness on miss trials was characterized by roughly equivalent responses for selecting the pre- or the post-change item’s attribute, which was modulated by gaze strategy during the trial. We will discuss how these results inform both visual change detection and higher-order theories of visual awareness.