Heat-Diffusion Approaches for 3D Computer Vision

Thursday, March 7, 2013 - 2:00pm - 3:00pm EST

  • Location:Dibner Building, LC400
    New York, US

Speaker: Dr. Yi Fang

Host Faculty: Professor Ramesh Karri

Abstract

Recent developments in data acquisition techniques have resulted in a rapid growth in the number of available three dimensional (3D) models across areas as diverse as engineering, medicine and biology. Researchers are regularly interested in interpreting the 3D shape of such models according to their intrinsic geometric attributes. The effective and efficient interpretation of 3D models is often challenged with the prevalence of non-rigidity within the shapes, the corruption of the shapes due to the presence of geometric noise, and the availability of a large volume of 3D models in innumerable databases. My work is concentrated on the development of a novel framework for 3D shape analysis, such as shape matching, segmentation, and retrieval, based on the effective utilization of the heat diffusion concept. The novelty of this framework is derived from an analogy between the process of 3D shape interpretation and that of heat transfer. The approaches exploit the intelligence of heat as a global structure-aware message that traverses across a meshed surface and is capable of exploring the intrinsic geometric features of the shape. I have demonstrated the performance of several heat-driven approaches for efficient non-rigid 3D shape registration, robust segmentation of 3D models, and efficient retrieval of 3D models with applications in engineering, medicine and biology. The experimental results indicate that heat-driven approaches are able to reveal the interpretations of 3D shape in a highly robust fashion, independent of any reference to prior knowledge, and in a manner consistent to human perception. In addition, the heat-driven approaches are very general and have great potential for applications to a broad range of research fields, for example, medical image processing, biological networks, social networks and semantic analysis of documents.

About the Speaker

Yi Fang received his Bachelor's and Master's degree in Engineering from Xi’an Jiao tong University, China, in 2003 and 2006, respectively. He received his PhD degree in Engineering from Purdue University, West Lafayette, USA on December, 2011. He worked as research intern in Siemens Corporate Research on 3D medical image processing. He then joined Riverain Technologies, a leader and technology innovator in the health care industry and beyond, as a Senior Research Scientist. He is now a Senior Staff Scientist in the Department of Electrical Engineering and Computer Science at Vanderbilt University. His current research interests are in computer graphics, computer vision, image processing, machine learning and their applications to multiple disciplines as diverse as engineering, medicine, biology and social science. He has co-authored 16 refereed papers in journals and top tier conferences, of which he is the first author in 9. Some of his works have been widely reported by both national and international media, such as Purdue Newsroom, Sciencedaily and Yahoo!.