Content-Based Multimodality Image Class

Current military surveillance platforms carry a wide range of sensor systems to permit target imaging under a wide variety of operational and atmospheric conditions. These disparate image sensors provide markedly different depictions of the same target, placing severe demands on the image analyst and greatly complicating development of automated methods for target classification. Existing methods of automated image matching typically rely on some form of correlation or template matching performed on the image or in the feature space—a very difficult process for images obtained from multiple sensor modalities. We propose to develop a novel methodology for target classification and content-based image retrieval that incorporates novel feature extraction and target classification methods based on recent developments in computer vision. During the course of this project, we will develop a prototype version of the algorithms and demonstrate their performance on a variety of ISAR, SAR, and electro-optic images. We will also design an open architecture database that can be used for rapid, automated image retrieval.