Object recognition by hierarchically organized anarchy

Bosco Tjan


Bosco Tjan received his Ph.D. in Computer Science at the University of Minnesota under the supervisions of Gordon Legge and Dan Kersten of Psychology. He subsequently worked with Heinrich Buelthoff at the Max-Planck Institute for Biological Cybernetics in Tuebingen, Germany, and then with John Oliensis at the NEC Research Institute in Princeton, New Jersey. He joined the psychology faculty at the University of Southern California in January of 2001. His research interest includes various aspects of visual processing pertained to object recognition, scene percpetion, and reading.


Ideally, the best representation for an object-recognition task should depend on the task and the object set. Since the complexity of the decision space varies greatly between tasks and object sets, one would expect multiple representations be used by the visual system for the purpose of object recognition. Most conventional theories of object recognition assume that within a single visual-processing pathway only one form of representation is derived and used to recognize objects. Versatility of the visual system comes from having multiple visual-processing pathways, each supporting a particular class of objects or recognition tasks.

We propose and simulate a simpler and theoretically more constrainted alternative, capable of explaining the same set of data and more. A single primary visual-processing pathway, loosely modular, is assumed. Memory modules are attached to decision sites along this pathway. Object-identity decision is made independently at each site. A site's response time is a monotonic-decreasing function of its confidence regarding its decision. The system's response is the first-arriving response from any site. The effective representation of such a system, if determined behaviorally, can appear to be self-adapting and specialized for different tasks and stimuli (resembling a number of recent neuro-cognitive findings). This, however, is merely a reflection of the decisions being made at the appropriate level of abstraction.

Maintained by Philippos Mordohai