Incongruent Objects in Real-World Scenes Distort Visual Memory Recall

Previous work by Greene et al. (2015) has found perceptual deficits in description and classification for highly incongruent scenes (where an object did not match its surrounding scene context) compared to congruent scenes. However it is still unclear how visual memory for complex scenes is affected by incongruent objects, particularly during free recall. Using a drawing task to analyze the visual contents of memories (as in Bainbridge et al., 2018), here we evaluate the differences in visual memory recall elicited by incongruent objects in real-world scenes. 30 participants were eye-tracked while viewing 12 real-world scene stimuli, displayed with a congruent or incongruent object in the foreground. For example, a construction scene contained either a congruent construction sign, or an incongruent blender. Participants viewed an equal number of congruent and incongruent scenes, which were counterbalanced across participants. After stimulus viewing, they performed a distractor task to disrupt verbal working memory strategies. Then, participants were asked to draw as many images as they could recall, in as much detail as possible. On average, participants drew 8.4 out of 12 images from memory, with an average of 4.2 correct object/scene pairings. Eyetracking analyses revealed that incongruent objects affected the overall viewing patterns of the scenes. No significant differences in memory performance at the scene-level were found between incongruent and congruent images. However, participants made different object/scene binding errors – either remembering objects in isolation, or transposing objects in wrong scenes entirely. Significantly more errors were made with objects originally shown in incongruent scenes than congruent scenes. Differences were also found in drawing stroke order for incongruent versus congruent scenes. Ultimately, incongruent objects within scenes do not so much affect the memory for the scene background or the incongruent object, but how these features are bound together.