Abstract
Traditional views of visual search have pointed to the importance of the presence or absence of certain basic "features" such as oriented or colored bars. Intuitions we have gained from running our recent model of search (Rosenholtz et al, 2012) suggest instead the primacy of concepts like tiling and coherence, as well as the less pithy insensitivity of a bag of derivatives to sign of contrast. Here we redesign five classic search experiments in ways that manipulate these latter factors while minimally changing the presence or absence of classic basic features: we take the bars making up classic search stimuli – Ts, Ls, Qs, etc. – and move them, or change their thickness, texture, or shape. We find search efficiencies of these slightly changed new tasks are significantly different from those of the classic ones, and not always in the way we expected. By only changing the thickness of the lines, we could make the O among Q task considerably easier (target-present search slope changed from 20.1ms/item to 13.1ms/item). In search for a T among Ls of two orientations, changing the horizontal bar in each symbol to a triangular "flag" made search less efficient (target-present search slope changed from 17.7 ms/item to 31.3 ms/item). We could make an easy Q among O task harder by simply moving the vertical line to a different part of the Q. Search efficiencies of classic and new conjunction tasks and tiled among vertical search tasks are also significantly different. These results highlight the importance of considering more image-based, less thing-based features in search, and the utility of model-driven stimulus manipulations.
Meeting abstract presented at VSS 2014