Sackler / PEB Invited Speaker

Prof. Jun Zhang
Event time: 
Friday, December 12, 2014 - 2:00pm
Location: 
Bass Center, Room 305 See map
266 Whitney Ave.
New Haven, CT 06511
Event description: 

Prof. Jun Zhang from the University of Michigan will be visiting and giving a 2 pm seminar on “Motion Sensors, Orientation Sensors, and Object Segregation.”

 

Abstract:

Jun Zhang (Professor of Psychology and Professor of Mathematics, The University of Michigan, Ann Arbor)

Object segregation is the process by which adjacent locations of the visual space are perceptually grouped into a “chunk” of connected region representing foreground (“object”) against an ambient background. This grouping/binding captures an object’s momentary location and the region of space it occupies. An example is the random-dot kinematogram (Braddick, 1974) in which the segregation of a camouflaged surface patch arises out of 2-D rigid translational motion cue alone, with no form information specified a priori. In this talk, I will present a model of motion-based figure-ground segregation through the interaction of motion sensors with orientation sensors. Responses of motion sensors, modeled as Reichardt detectors computing spatial-temporal image correlation, confound image structure with velocity information (“aperture problem”), and hence cannot immediately provide the a basis for binding. The problem is solved by resorting to a differential geometric framework, whereby motion sensor responses are modeled as tangent vectors, and orientation sensors supply the affine connection information governing the comparison of adjacent tangent vectors. Geometric quantities, such as Riemannian metric, Levi-Civita connection, geodesic, etc, can all be implemented through the network of orientation sensors with lateral interactions both within preferred tunings and across locations. Object segregation results from direct ”readout” of motion sensor responses under the gating of orientation sensors, thereby avoiding the chicken-egg problem of extracting velocity prior to object segregation. Finally, for biological plausibility, I present a method of separating orientation tuning and directional tuning in single neuron activities that may reflect an interaction between orientation and motion pathways.

Click here to see the flyer.