|
|
The Departments of Computer and Information Science
and
FCRH Interdisciplinary Faculty Seminar in
BioInformatics and Big Data
Present
Dr. Daniel Leeds
Department of Computer and Information Science
Fordham University
Title: Exploring visual feature spaces for cortical object perception
Date: Wednesday, January 29, 2014.
Time: 1:00 p.m.
Place: Room JMH 302
Abstract:
Object perception employs a network of brain regions that encodes a hierarchy of increasingly complex visual features. While the earliest stages of human vision have been reasonably well-modeled, the visual properties used in higher-level brain regions is less well understood. Although prior studies have identified individual shapes exciting selected neurons in mid-level vision, they provide little insight as to the principles used to distinguish these preferred stimuli. Here, we explore the grouping of properties salient to cortical perception by defining and using Euclidean feature spaces based on complex visual properties of natural and synthetic object classes. Natural "real-world" object images are projected into a space reflecting visual grouping based on a SIFT bag-of-words distance metric (Nowak 2006). Synthetic "Fribble" objects (Williams 2000) --- animal-like objects composed of geometric shapes --- are projected into a space capturing the morphs between Fribble appearances. Studying fMRI data in the context of these novel feature spaces, we observe that object stimulus pairs near one another in "SIFT" and "Fribble" spaces can produce responses at the opposite extremes of the measured activity range for pre-selected brain regions in the visual object perception pathway. These observations may demonstrate extension of surround suppression observed in lower levels of vision --- such as the Gabor-like models of V1 receptive fields (Kay 2008, Hubel and Wiesel 1962).
Bio:
Dr. Daniel Leeds is an assistant professor of Computer and Information Science at Fordham University. Dr. Leeds' research focuses on models of biological perception, developing techniques in machine learning, signal processing, and computer vision. His work studies the statistics of neural data, of human behavior, and of the visual world. Dr. Leeds received his PhD in Neural Computation at Carnegie Mellon University in 2013, where he was an RK Mellon Foundation Presidential Fellow in Life Sciences and a National Science Foundation Graduate Research Fellow.
For directions and information, please contact Ms. Palma Hutter at hutter@fordham.edu or 718-817-4480.
|
| |
|
|
|
