Boundary extension (BE) is a remarkably consistent visual memory error in which participants remember seeing a more wide-angle image of a scene than was actually viewed (Intraub & Richardson, Journal of Experimental Psychology: Learning, Memory, and Cognition 15:179-187, 1989). Multiple stimulus factors are thought to contribute to the occurrence of BE, including object recognition, conceptual knowledge of scenes, and amodal perception at the view boundaries (Intraub, Wiley Interdisciplinary Reviews: Cognitive Science 3:117-127, 2012). In the present study, we used abstract scenes instead of images of the real world, in order to remove expectations based on semantic associations with objects and the schematic context of the view. Close-angle and wide-angle scenes were created using irregular geometric shapes rated by independent observers as lacking any easily recognizable structure. The abstract objects were tested on either a random-dot or a blank background in order to assess the influence of implied continuation of the image beyond its boundaries. The random-dot background conditions had background occlusion cues either present or absent at the image border, in order to test their influence on BE in the absence of high-level information about the scenes. The results indicate that high-level information about objects and schematic context is unnecessary for BE to occur, and that occlusion information at the image boundary also has little influence on BE. Contrary to previous studies, we also found clear BE for all conditions, despite using scenes depicting undetailed objects on a blank white background. The results highlighted the ubiquitous nature of BE and the adaptability of scene perception processes.

Studies have shown that people consistently remember seeing more of a studied scene than was physically present (e.g., Intraub & Richardson Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 179-187, 1989). This scene memory error, known as boundary extension, has been suggested to occur due to an observer's failure to differentiate between the contributing sources of information, including the sensory input, amodal continuation beyond the view boundaries, and contextual associations with the main objects and depicted scene locations (Intraub, 2010). Here, "scenes" made of abstract shapes on random-dot backgrounds, previously shown to elicit boundary extension (McDunn, Siddiqui, & Brown Psychonomic Bulletin & Review, 21, 370-375, 2014), were compared with versions made with extremal edges (Palmer & Ghose Psychological Science, 19, 77-84, 2008) added to their borders, in order to examine how boundary extension is influenced when amodal continuation at the borders' view boundaries is manipulated in this way. Extremal edges were expected to reduce boundary extension as compared to the same scenes without them, because extremal edge boundaries explicitly indicate an image's end (i.e., they do not continue past the view boundary). A large and a small difference (16 % and 40 %) between the close and wide-angle views shown during the experiment were tested to examine the effects of both boundary extension and normalization with and without extremal edges. Images without extremal edges elicited typical boundary extension for the 16 % size change condition, whereas the 40 % condition showed signs of normalization. With extremal edges, a reduced amount of boundary extension occurred for the 16 % condition, and only normalization was found for the 40 % condition. Our findings support and highlight the importance of amodal continuation at the view boundaries as a component of boundary extension.

No abstract

No abstract

Individuals consistently remember seeing wider-angle versions of previously viewed scenes than actually existed. The multi-source model of boundary extension (BE) suggests many sources of information contribute to this visual memory error. Color diagnosticity is known to affect object recognition with poorer recognition for atypically versus typically colored objects. Scenes with low-color diagnostic main objects and two versions of scenes with high-color diagnostic main objects (typically and atypically colored) were tested to determine if the reduced ability to identify the main object in a scene influences BE. Scenes were presented to one group of participants for 46 ms and another group for 250 ms. Each scene was followed by a mask and a request for a recognition response concerning the identity of the main object. The scene was then immediately presented again for testing and participants rated it as depicting a more close-up view, more wide-angle, or the same view as before. The study demonstrates that poorer encoding of main objects in scenes leads to increased BE, but trial-by-trial recognition accuracy had no relationship to BE magnitude. This finding provides further insight into the impact of task demand and main object recognition on BE.