Abstract
Purpose: The fidelity of color memory is controversial. Here we compared directly the bias and precision of color perception and memory for real objects. Method: Observers viewed 16 painted wooden cubes in two matte gray booths and made color matches by selecting the best-matching paint from a commercial booklet of 985 samples. The matching booklet was either under the same or different illumination as the cube and matches were made either simultaneously (Perception) or after a 15-minute delay (Memory). Each observer made matches for four cubes in each of the four conditions, and cubes were counterbalanced across conditions. Cube colors evenly sampled the chromaticity space defined by the 985 paint samples (u’v’ space). Results: 1. Bias: For each cube, we computed the average chromaticity of the paint samples chosen by observers in each condition. Bias was defined as the difference between average matches in perception and memory. A permutations test showed that few of the biases were significant. Furthermore, the size of observed biases is consistent with a simple ideal observer model that takes into account the discretization of the matching booklet and formulates memory as unbiased but less precise than perception. 2. Precision for each cube was the average Euclidean distance between each chosen paint sample and the average match (in u’v’) in that condition. Memory for cube colors was significantly less precise than was perception for the same colors (p<0.05). In addition, across cubes, significantly more paint samples were chosen in memory conditions than in perception conditions (p<0.005). The illumination condition had no effect on number of paint samples chosen (p=0.45). Conclusions: The data suggest that memory for real, illuminated cubes is less precise than perception. However, under these experimental conditions, memory is not systematically biased compared to perception.
Meeting abstract presented at VSS 2012