Jasmin Leveille, Stephen Grossberg and Massimiliano Versace

Department of Cognitive and Neural Systems, Center for Adaptive Systems, and Center of Excellence for Learning in Education, Science, and Technology, Boston University, Boston, MA 02215

Various motion percepts have been shown to obey a rule of vector decomposition, where global motion appears to be subtracted from the true motion path of localized stimulus components (Johansson, 1950). This results in striking percepts wherein objects and their parts are seen as moving relative to a common reference frame. While vector decomposition has been amply confirmed in a variety of experiments, no model to date explains how it may occur in neural circuits. The current model shows how vector decomposition results from V1/V2/MT interactions postulated in the motion stream of the 3D Formotion model (Grossberg, 1991; Francis and Grossberg, 1996; Baloch and Grossberg, 1997; Grossberg, Mingolla and Viswanathan, 2001; Berzhanskaya, Grossberg and Mingolla, 2004). In order to assign ambiguous motions to their correct depths, the model predicts how motion signals at a nearer depth may be subtracted from corresponding signals at a farther depth. In response to displays wherein vector decomposition is possible, this results in a shift in motion direction. This mechanism has enabled us to simulate classical Duncker (1929) and Johansson (1950) percepts. These simulations clarify the interplay of mechanisms of depth and motion perception in the formation of a frame of reference.

Supported in part by the NSF (SBE-0354378) and DARPA (HR001109-03-0001, HR001-09-C-0011).