Fundamental to visual object recognition is the ability to recognize abstract categories of objects (e.g., cups versus pens) as well as specific exemplars within those categories (e.g., individual pens). Interestingly, this ability poses a dilemma for the visual system: How can it recognize that two shapes should be considered the same (i.e., belong to the same abstract category) yet also different (i.e., correspond to different exemplars)? The aim of the proposed work is to investigate how the human brain may implement a solution to this problems. In particular, the goal is to uncover the architecture of functionally defined, neurally dissociable subsystems that underlie object recognition, and to determine whether this architecture reflects a solution to the abstract/specific dilemma. Understanding the structure of component subsystems has primary importance, as it must be addressed before satisfactory answers can be offered for contemporary questions in this field; different answers may apply to different subsystems. Preliminary studies indicate that dissociable subsystems learn to operate in parallel to accomplish abstract-category and specific-exemplar recognition of visual objects. However, it is difficult to produce fail-safe dissociations of functionally specified subsystems, and other architectures remain viable as alternative theories (e.g., dissociable subsystems operating in sequence, a single general-purpose mechanism, attention to different information within a single subsystem). Thus, the proposed research will further test and refine these theories, using a converging evidence attack to draw strong conclusions. The research will integrate analyses of the goals of the visual system with evidence of the neural implementation of dissociable subsystems to constrain such theories. Divided-visual-field studies will test whether abstract-category and specific-exemplar recognition subsystems operate in parallel (rather than in sequence) and with different relative efficiencies in the left and right cerebral hemispheres. They also will test the particular levels of categorization performed by abstract and specific subsystems, whether stimulus and task demands influence the relative contributions of these subsystems in predictable ways, as well as whether these subsystems utilize contradictory processing strategies (e.g., features-based versus whole-based processing). Overall, should evidence for dissociable parallel subsystems be observed, object recognition theories that attempt to account for performance through different architectures would have to be significantly revised. In any case, this research should lead to a greater understanding of the component subsystems underlying visual object recognition, with implications for addressing why neurological damage can produce selective visual recognition impairments and for suggesting useful architectures in computer vision systems.
Showing the most recent 10 out of 11 publications
