Abstract
Humans can view and remember thousands of natural images with a remarkable degree of accuracy (Brady, Konkle, Alvarez & Oliva, 2008). However, memory for simpler stimuli is much less robust. What image parameters are required to support massive memory capacity? Here we present an initial exploration for three types of images: colored patches, image textures, and object photographs. Stimuli were presented on a color-calibrated monitor. Color patches were uniform, and were chosen by: (1) identifying 16 points that sampled uvY color space and (2) choosing chromaticity values sampled equidistantly from concentric rings around each of the 16 points. Color patches were clearly discriminable from each other. Image textures were close up color images of real-world objects chosen to lack obvious cues to object identity or scene, such as clothing swatches. Photographs were black and white versions of objects shown on backgrounds. All images were identically sized and subtended ~ 5 degrees of visual angle. Task: For each image type, observers (n = 25) completed an exposure task and a memory task. During the exposure phase, observers viewed exemplars for one second each and instructed to remember them for later testing. The memory task was a 2AFC task in which observers were simultaneously shown one familiar image and one novel image of the same type. Observers chose the novel image on each trial. Results: Performance for objects was similar to performance previously reported for colored natural images (.93 +/- .01). Performance for colored patches was not different from chance (0.53 +/- .02), and performance for textures was intermediate (0.83 +/- .02). Reaction times for natural images were significantly faster than for colored patches or textures, and there was no difference between colored patches and textures. These data provide some constraints on the mechanisms subserving visual memory capacity.
Meeting abstract presented at VSS 2013