The role of spatial frequency in expert object recognition.

Abstract

Novices recognize objects at the basic-category level (e.g., dog, chair, and bird) at which identification is based on the global form of the objects (Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976). In contrast, experts recognize objects within their domain of expertise at the subordinate level (e.g., Sparrow or Finch) for which the internal object information may play an important role in identification (Tanaka & Taylor, 1991). To investigate whether expert recognition relies on internal object information, we band-pass filtered bird images over a range of spatial frequencies (SF) and then masked the filtered image to preserve its global form. In Experiment 1, bird experts categorized common birds at the family level (e.g., Robin or Sparrow) more quickly and more accurately than novices. Both experts and novices were more accurate when bird images contained the internal information represented by a middle range of SFs, and this finding was characterized by a quadratic function in which accuracy decreased toward each end of the SF spectrum. However, the experts, but not the novices, showed a similar quadratic relationship between response times and SF range. In Experiment 2, experts categorized Warblers and Finches at the more specific, species level (e.g., Wilson's Warbler or House Finch). Recognition was again fastest and most accurate for images filtered in the middle range of SFs. Collectively, these results indicate that a midrange of SFs contain crucial information for subordinate recognition, and that extensive perceptual experience can influence the efficiency with which this information is utilized. (PsycINFO Database Record

Neural and behavioral effects of subordinate‐level training of novel objects across manipulations of color and spatial frequency

“…That is, when LSF information was removed from images of faces, the N170 resembled the response to objects and face-specific N170 effects were not present. Combined with the present results, as well as the report by Hagen et al (2016), these findings suggest that whereas expert-level face processing may rely more on LSF information, learning a new object category and acquiring object expertise may require increased use of mid-and possibly high-range spatial frequency information.…”

“…In addition, although accuracy (d') differed across image manipulations (performance greater for color than grayscale/HSF/LSF images and for grayscale/HSF relative to LSF), these differences were not impacted by training (pre, post) or training condition (subordinate, basic). Overall, it is possible that the superiority of mid-range SF information demonstrated by Hagen et al's (2016) experts during species level identification was captured in the present study by faster response times for HSF images following training. However, it is difficult to directly compare these two studies for several reasons.…”

“…First, the present investigation only examined the early stages of the acquisition of expertise (first 2 weeks) whereas Hagen et al examined real-world experts with years of birding experience. Second, Hagen et al (2016) used bird images and a category verification task and we used novel objects and a serial matching task. Third, Hagen and colleagues did not test novices with a species-level identification task (the equivalent of our subordinate-level training), as performance would likely be very poor and so it is unclear whether novices would also rely on mid-range spatial frequencies during a species-level identification/discrimination task.…”

“…Third, Hagen and colleagues did not test novices with a species-level identification task (the equivalent of our subordinate-level training), as performance would likely be very poor and so it is unclear whether novices would also rely on mid-range spatial frequencies during a species-level identification/discrimination task. Finally, as described above, the implementation of the SF conditions was different resulting in our HSF condition containing both the mid-SF and high SF ranges reported by Hagen et al (2016).…”

“…Subordinate-level, but not basic-level, training effects were expected to generalize to untrained exemplars within trained object families. In addition, based on previous findings suggesting that real-world experts differentially utilize color and spatial frequency information (Hagen et al, 2014(Hagen et al, , 2016, it was predicted that performance and neural responses would be disrupted by color and spatial frequency image manipulations after subordinate-level training. Finally, based on previous work reporting links between the neural responses associated with perceptual expertise and behavior measures (Gauthier et al, 2003;Tanaka & Pierce, 2009) exploratory analyses examined the extent to which significant brainbehavior correlations were present in this study.…”