Sparse Representation of Multimodality Sensing Databases for Data Mining and Retrieval

We propose to apply recently developed methods of sparse representation and dimensionality reduction to multimodality image and video databases. Our research will consist of three interconnected components 1 multimodality feature extraction from the database 2 informationtheoretic similarity measures for pairwise matching 3 hierarchical similarity-based clustering and database updating. Information-theoretic measures, sparse approximation and dimensionality reduction will play key roles in our work. They will allow us to reduce complexity, accelerate query matching times, improve specificity of the query matches, and incorporate robustness to noise and other distortions. Experimental validation will be performed by a combination of simulation and experiment on multimodality databases. As part of this proposal we propose to build a small scale experimental LADAREO video acquisition testbed.