Purchase this article with an account.
Simen Hagen, Quoc C. Vuong, Lisa S. Scott, Tim Curran, James Tanaka; The role of spatial frequencies in expert object recognition. Journal of Vision 2014;14(10):1287. doi: 10.1167/14.10.1287.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Objects are typically recognized at the basic level (e.g., bird) in which the external contour shape of the object is important for recognition. Experts, on the other hand, typically recognize objects at the subordinate level (e.g., sparrow), which seem to depend more on internal features of objects. Here we investigated whether bird experts rely on internal object features to facilitate fast and accurate subordinate-level recognition of objects in their domain of expertise. We filtered bird images over a range of spatial frequencies corresponding to 2-4 cycles per image (cpi), 4-8 cpi, 8-16 cpi, 16-32 cpi, and 32-64 cpi. This manipulation preserved the external shape of the object while systematically degrading its internal feature information. In Experiment 1, bird experts and novices categorized common birds at the subordinate, family-level (e.g., robin, sparrow, cardinal). The main finding was that experts were faster and more accurate than novices. Moreover, experts were fastest with images filtered between 8 cpi and 32 cpi in which external and internal information were both preserved. In Experiment 2, experts categorized birds at the subordinate, species level (e.g., Wilson's warbler, Tennessee warbler), in which external shape is less diagnostic and internal features are more important to identification. For species-level categorizations, experts were faster at recognizing birds filtered between 4 cpi and 32 cpi relative to images filtered at 2-4 cpi. Response time distribution analyses revealed that only images filtered at 8-16 cpi led to a reaction time advantage in the fastest trials. In summary, bird experts form elaborate object representations that include information about the internal features of birds allowing them to make fast and accurate subordinate-level recognition for objects in their domain of expertise.
Meeting abstract presented at VSS 2014
This PDF is available to Subscribers Only