Abstract
Estimating object properties from incoming sensory information is challenging because of the many-to-one mapping of object properties to sensory signals. Here, we study how sensory ambiguity interacts with memory in perceptual estimates of surface lightness. Observers compared two briefly presented 2-degree patches and indicated which one appeared lighter. The patches were presented either simultaneously or with a 2.5-second inter-stimulus interval, either against identical backgrounds or backgrounds differing in luminance. This 2x2 design allowed us to study how perceived lightness is affected by memory (delay), context (differing backgrounds), and both combined. We fit psychometric functions to the proportion-lighter data in each condition to estimate both bias and precision. Memory, but not context, decreased precision. Both memory and context change induced biases, and these biases were not additive. Probabilistic models can explain perceptual biases as arising through a combination of noisy sensory evidence with prior information about the world. We developed two such models: One model bases lightness judgments on inferred reflectance, the other on edge contrast. Priors were developed from the composition of experimental stimuli. In each model, one set of parameters was used for all experimental conditions. Both models produced biases due to memory and context, but the reflectance model provided a better quantitative account. Only the reflectance model accounted for the non-additivity of biases in the combined condition. We conclude that memory for lightness is biased, and that memory interacts with perceptual processing. This interaction is consistent with a model where lightness judgments rest on inferred reflectance.