Extracting Soft Biometric Text Descriptors from Images of Mass Disaster Victims
Brown Bag Lecture by Niyati Chhaya | 3/13/2012 11AM-12PM | 7th Floor Conference Room, Bldg 38A
Abstract: Earthquakes, hurricanes, terrorist attacks, and other such events can cause tremendous harm to people and infrastructure, often leaving those outside the areas impacted with little or no information on the state of their friends and relatives who may have been affected. Such situations create a need to provide information about the injured and the missing to the public from various emergency medical care centers. This research aims at utilizing patient triage images captured by the Lost Person Finder for extracting soft-biometric information to construct text-based descriptors, making it easier to search for someone without any privacy issues.
This work explores various approaches to extract different appearance-related features from person triage images and establishes the need to exploit relationships between different features to assure accuracy in the text description. We show how interaction between individual feature detectors can lead to increased accuracy in the resulting text descriptor using a Markov Network. As the main aim of this work is to search individuals, an Amazon Mechanical Turk survey is also presented to understand how people describe others. We show how text from human annotations indicates relationships between various features and descriptions. We then use of probabilistic graphical model to improve and correct a computer vision task of feature extraction supported by data from natural language processing.
Bio: Niyati Chhaya is a PhD candidate in the Department of Computer Science and Electrical Engineering at the University of Maryland, Baltimore County (UMBC). Her dissertation research, with Dr. Tim Oates, is focused on applying probabilistic graphical models for combining computer vision with natural language processing where she is working on building a text representation to exploit relationships between different soft biometric features. This research is based on the Lost Person Finder project at NLM, NIH. Her research interests include machine learning, computer vision, and natural language processing. She holds a Masters’ Degree from UMBC and an undergraduate degree in Computer Science from University of Pune, India.