October 9 - 12, 2012
Building 10 & Natcher Conference Center
- General Schedule of Events
- Opening Plenary Session
- Concurrent Symposia Sessions
- Poster Sessions
- National Graduate Student Research Conference
- FARE Award Ceremony
- Special Exhibits on Resources for Intramural Research
- TSA Research Festival Exhibit Show
- Clinical Center Tours
- Research Festival Committees
- Past Research Festivals
FaceMatch: visual search by photos of missing persons during a disaster event
| Thursday, November 07, 2013 — Poster Session II | |||
|---|---|---|---|
12:00 p.m. – 2:00 p.m. |
FAES Academic Center (Upper-Level Terrace) |
NLM |
RSCHSUPP-6 |
Authors
- E. Borovikov
- S. Vajda
- G. Lingappa
- S. Candemir
- S. Antani
- M. Gill
- G. Thoma
Abstract
NLM's PeopleLocator (PL) is a Web-based system for family reunification in cases of natural or man-made disasters. PL collects photos and brief text meta-data (name, age, etc.) of missing or found persons. Currently supported text queries may be insufficient because text data are often incomplete or inconsistent. Adding an image search capability would significantly benefit the user experience. We report on our face matching R&D that aims to provide robust face localization and matching on digital photos of variable quality. We also present our approach to image near-duplicate image detection. Face localization is done via skin-tone/landmarks enhanced grayscale face detector, more accurate than many open source and commercial detectors. Face matching is done via an ensemble of image descriptors (HAAR, SIFT, SURF, ORB), using a smart re-ranking procedure. We describe the integration of our face matching system with PL, report on its performance, and compare it to other publicly available face recognition systems. In contrast to these systems that have many good quality well-illuminated sample images for each person, our algorithms are hampered by the lack of training examples for individual faces, as those are unlikely in a disaster setting.
