Abstract:
Performance evaluation is a key requirement in the research and development of pattern recognition techniques. Ongoing success depends on our ability to gauge progress. While various measures have gained widespread acceptance over the years, they sometimes fail to capture important aspects of a system and its utility in the real world. In this talk, I will discuss a number of issues that arise when attempting to evaluate the performance of document analysis systems, a rich domain that involves tasks ranging from low-level image processing steps all the way up to high-level language understanding problems. I will also present some new ideas for evaluating the performance of such systems that offer significant benefits in terms of soundness and reproducibility.

Bio:
Daniel Lopresti is Professor and Chair of Lehigh's Department of Computer Science and Engineering, as well as Director of the university's new Data X Strategic Initiative. He conducts research examining fundamental algorithmic and systems-related questions in pattern recognition, document analysis, and computer security. Dan received his Bachelor’s degree from Dartmouth College in 1982, and his Ph.D. in computer science from Princeton University in 1987.

Dan arrived at Lehigh in 2003 after having spent time on the faculty at Brown and on the research staffs at Panasonic Technologies and Bell Labs, Lucent Technologies. He has published over 150 papers in professional conferences and journals, and he holds 24 U.S. Patents. Dan has held leadership roles in most of the major international conferences on document analysis over the years. He previously chaired Technical Committee 11 of the International Association for Pattern Recognition, and currently serves as chair of the IAPR’s Conferences & Meetings Committee.