Drexel Vision Lab Publishes Four Papers in European Conference on Computer Vision

Geoff Oxholm
Oxholm with one of the posters he displayed at the conference.

December 12, 2012 —  

The Drexel Vision Lab, lead by Dr. Ko Nishino, associate professor of computer science, published four papers in the top-tier European Conference on Computer Vision. The conference, which meets every other year and has an acceptance rate of 20 percent, is among the top three Computer Vision conferences in the world.

Two of the accepted papers, “Shape and Reflectance from Natural Illumination,” by Geoff Oxholm, CS Ph.D. candidate, and Dr. Nishino, and “Reflectance and Natural Illumination from a Single Image,” by Stephen Lombardi, CS Ph.D. candidate, and Dr. Nishino, investigate the inference of physical properties of the world that are captured within a photograph. “Shape and Reflectance from Natural Illumination,” describes an algorithm for learning an object’s shape and material appearance properties (e.g., its color, glossiness, dullness, etc.) using only an image of that object and an image of the world surrounding the object. Similarly, “Reflectance and Natural Illumination from a Single,” describes an algorithm for recovering the material appearance properties of an object and an image of the world surrounding the object given an image of that object and its shape.

The novelty of these methods lies in the fact that they operate on photographs taken under arbitrary illumination conditions, whereas much of the previous work assumes the input photographs are taken in laboratory “dark room” settings with carefully controlled illumination. This is an important step for a number of applications (including material recognition, object tracking, and others) because it allows for the identification of material appearance properties in ordinary, everyday photographs. The database used in these papers was released at the conference. Software for the second paper was also released.

The third work, entitled, “The Scale of Geometric Texture” by Oxholm, Prabin Bariya, a recent CS master’s graduate, and Dr. Nishino, is a new take on the study of texture in images. Rather than study the visual properties of texture, it looks at surface geometry’s role in texture appearance. The database was released at the conference.

The fourth work, entitled, “Going With the Flow: Pedestrian Efficiency in Crowded Scenes,” written by Dr. Louis Kratz, recent CS Ph.D. graduate, and Dr. Nishino, shows how measuring a pedestrian’s conformance with the crowd around them leads to more accurate tracking, and unusual event detection.

Oxholm, Lombardi, and Dr. Nishino attended the conference to present their work. View the video of Lombardi's talk.