Abstract
Visual search with homogeneous distractors has been extensively characterized within a modeling framework, both with signal detection theory (Palmer et al, 2000), and optimal-observer models (Mazyar et al, 2012). Visual search with heterogeneous distractors has also been investigated, starting with Duncan and Humphreys (1989), but a much smaller proportion of this work used simple parametric stimuli, making modeling and dissociation of component processes more difficult. Here, we performed a detailed characterization of the factors that influence performance in visual search with heterogenous distractors: set size, task (detection vs localization), temporal order of revealing target identity and search array (perception vs memory conditions) and stimulus spacing (distant vs nearby stimuli). We captured this data with optimal-observer models and quantified the influence of these factors on model parameters. In both tasks and conditions, performance decreased with set size, and also as the most similar distractor was closer in orientation distance to the target. In contrast, increased proximity in orientation distance to the mean of the distractors did not decrease performance and neither did increases in the heterogeneity of the distractors. These patterns of results were captured by an optimal-observer model with a variable precision encoding stage. The fitted mean precision parameters decreased with the set size of the array and were higher in perception than in memory in both tasks. Adding a decision noise parameter improved the model fits on the detection data, but not on the localization data. Additionally, we were able to capture both the localization and detection data with a model with joint encoding parameters, suggesting that observers might be using the same encoding processes across both tasks. We replicated our results in a separate experiment with reduced stimulus spacing. These results futher demonstrate the value of using parametric stimuli and optimal-observer modeling to better understand visual search.
Acknowledgement: R01EY020958