COMPUTER VISION AND PATTERN RECOGNITION GROUP
Humans and animals use their visual abilities to navigate the world, forage for food, and survive. Is it possible to replicate some of these abilities on a computer so that they can assist us and enhance our quality of life by being active components of our day to day life? This is next generation science and technology. In this general context, the computer vision and image analysis group is actively engaged in designing intelligent computer algorithms to extract information such as 3D structure, 3D motion, object or person identity, material property, or geometric shape from images. It prides itself in insisting on tackling contemporary problems such as those in medical image analysis and marine sciences and on addressing difficult, still unsolved, fundamental problems in areas such as document processing, information extraction from video, perceptual organization, evaluation, non-rigid object modeling, and movement analysis.
- Faculty: Rangachar Kasturi, Dmitry B. Goldgof, Sudeep Sarkar
- Affiliates: Kevin W. Bowyer, Horst Bunke, Jeffrey Krischer, Lawrence O. Hall, and Thomas Sanocki
- Researchers: Post Doctoral Researcher, 12 Ph.D. Students, eight Masters students, and four undergraduates
- Graduates: More than 25 Ph.D. students have graduated from the lab and placed in universities and leading research labs both in the government and the industry.
Funding sources include the U.S. National Science Foundation, U.S. Army, Office of Naval Research, Defense Advanced Research Projects Agency (DARPA), Advanced Research and Development Activity (ARDA), National Institutes of Health, and The Whitaker Foundation.
VIDEO SURVEILLANCE PERFORMANCE MEASURES AND BENCHMARKING: Automated video surveillance is of great interest. This project is about evaluating object detection and tracking algorithms for video surveillance so as to benchmark progress and to identify areas for improvement and research. The specific objects of interest are face, text, hands, person and vehicle in video. Along with Advanced Interfaces, Inc., we are in the process of building the largest evaluation dataset to date (about 31 hours of video). This project was funded by Advanced Research and Development Activity (ARDA).
PLANKTON CLASSIFICATION: The goal of this project is to provide marine scientists with the ability to rapidly determine the plankton composition in a region of water. Normally this process would be a painstaking and tedious task, but the SIPPER software, developed by the Vision Group, makes use of active learning techniques to aide the scientist in classifying thousands of plankton images in a relatively short period of time. The SIPPER software package is an image classification system intended to identify plankton images that were generated by the SIPPER device. This is a joint project with the College of Marine Science.
BIOMETRICS: NEAR AND FAR, INDOOR AND OUTDOOR: Biometrics refers to the ability to recognize and verify the identity of persons based on physical characteristics, such as fingerprint, face, iris pattern, voice, ear shape, and walking style (gait). Applications are in secure communication, access control, security, and surveillance. We have been active in looking at biometrics, both for near and far (at-a-distance) types of scenarios, in indoor and outdoor conditions. Our impact has been in establishing multi-modal standard datasets and challenging problems for identification at a distance using (2D + 3D) face, ear, and gait. This line of research was funded by DARPA, U.S. Army and NSF-I/UCRC at USF.
AUTOMATED SIGN LANGUAGE RECOGNITION: This project is focused towards designing computer vision based, automated sign language recognition algorithms, so as to enhance communication with deaf persons. Sign languages are complex, abstract linguistic systems, with their own grammars, and their "articulation" involve not just the hands, but the face, shoulder, and arms. The Vision Group is pushing the state-of-the-art in scalable formalisms to recognize signs in sentences, purely from visual input, without the use of special equipment such as data gloves or magnetic markers. This research was being funded by the U.S. National Science Foundation (ITR) and is being conducted with faculty from the Department of Special Education at USF.