Multiscale Feature Extraction and Matching for Content-based Multi-modality Sensor Database Retrieval

Rapid, accurate target identification is a key component of force protection and accurate munitions delivery in modern warfare. However, the wide range of imaging modalities, from EO/IR sensors to SAR/ISAR sensors can present significant challenges and strains on the warfighter tasked with target identification. Rapid automated target identification methods that can narrow the scope of possibilities would be a major boon in reducing requirements on the warfighter, improving his efficiency. However, reliable feature matching across multiple sensing modalities presents novel challenges. For robust, reliable multi-modality image searching and matching, we must bring several technologies together. Multiscale feature matching methods can help accelerate the matching process, while novel statistical metrics can significantly aid cross-modality matching and improve robustness to extended operating conditions. It may be necessary to not only match hierarchically against coarse feature representations and the full feature set, but also to match the observed 2D imagery against 3D target Scientific Systems Company, Inc. is proposing an innovative multiscale, multi-modality image database methodology. Our approach incorporates: multiscale matching technology inspired by the Fast Multipole Methods exploited for rapid physics simulations of multi-body electromagnetics and gravitation, statistical matching metrics to enhance invariance to scaling, illumination, resolution, occlusion and other sources of variability, and methodologies developed by the Principal Investigator, among others for robust multiscale, structure-based, cross-modality association and registration. We believe our approach will significantly aid the warfighter in rapidly identifying and eliminating potential threats, relieving him of much of the burden of target identification.