The goal of this CAREER project is to develop an interdisciplinary research and education program for investigating the underlying theoretical and computational principles of machine-learning-based image annotation and retrieval. The novel research approach uses an automated learning-based image annotation method using a 3D-Hidden Markov Model (HDD) stochastic model that directly uses pixel level data as opposed to using the results of a segmentation algorithm. Using low-level features to learn high-level semantics is an important step towards automatic annotation of images. The research of this project focuses on three areas: (1) developing highly efficient machine learning mechanisms for imagery data, (2) developing models for high-dimensional imagery, and (3) developing a learning-based image annotation and retrieval system. This research will fundamentally improve image annotation technologies. This advance will provide theoretical understanding to the problem of managing and interpreting imagery data. The results of this research will also have impacts beyond pattern recognition for image databases. Many of these results can also be applied to other machine learning and data mining problems. This research will gain deeper insights into the principles of image understanding and annotation, and will ultimately contribute to the creation of intelligent and robust multimedia information systems. The educational plan includes developing interdisciplinary curriculum, and promoting diversity in the students' participation in this project. The scientific results of the research will enhance computer technology for recognizing objects and scenes with direct applications in online information management, homeland security, the military, and many scientific applications, including healthcare. Scientific publications and the project Web site http://riemann.ist.psu.edu will be used for the research results dissemination.
|Effective start/end date
|7/1/04 → 6/30/11
- National Science Foundation: $533,670.00