Computational Methods for High Resolution Imaging and Data Mining

Abstract : Research problems investigated during the funding period included enabling mathematics and computational research in integrated optical systems design as well as spectral data mining and analysis. Particular applications of this research to enhance image restoration of integrated optical-digital systems and space object identification were addressed. A novel minimization-maximization approach was proposed and successfully employed to optimize the design of integrated imaging systems that maximize the information strength in an incoming image signal before it is discretized and sampled in a CCD array. Five research papers were published on this work. Novel constrained nonnegative matrix factorization (CNMF) algorithms were developed and successfully used to extract material features and material composition from spectral traces of man-made space objects, such as geosynchronous satellites. A total of four papers were published as a result of this work. Technology transfer included collaboration with researchers from Oceanit Labs and Boeing Inc. (2 research papers) and transition of our CNMF algorithms for text mining and document clustering applications (2 research papers). Research results were presented at the 2004 SIAM International Conference on Data Mining, at the Annual SPIE meeting and at the AMOS Technical Conference in both 2003 and 2004.