Large-Scale Image Search

Short bio

Sung-Eui Yoon is an associate professor at Korea Advanced Institute of Science and Technology (KAIST). He received the B.S. and M.S. degrees in computer science from Seoul National University in 1999 and 2001 respectively. He received his Ph.D. degree in computer science from the University of North Carolina at Chapel Hill in 2005. He was a postdoctoral scholar at Lawrence Livermore National Laboratory.  His research interests include scalable graphics algorithms for rendering and proximity computations for massive models. He has published more than 50 technical papers in top journals and conference related to visualization and graphics. He also gave numerous tutorials on ray tracing, collision detection, and image search in premier conferences like ACM SIGGRAPH, IEEE Visualization, and CVPR. He served as conf. co-chair and paper co-chair for ACM I3D 2012 and 2013 respectively. At 2008, we published a monograph on real-time massive model rendering with other three co-authors. Some of his papers received a test-of-time award, a distinguished paper award, and a few invitations to IEEE Trans. on Visualization and Graphics.

Abstract

Large-scale image search is now considered as one of fundamental tools that are used in many data-driven computer vision and other related applications. Nonetheless, handling billions of images poses various technical challenges. In this talk, I will present 1) binary code embedding techniques of image descriptors, 2) methods of encoding quantization errors (or distance) within those binary codes, and 3) accurate shortlist computation that improves the accuracy of image search. I will also discuss other related techniques (ray tracing and collision detection) that we have been working together with image search in a broad field of proximity computing.