Nonlinear Enhancement of Visual Target Detection

Object identification, quantification and localization are key goals in both military target detection and quantitative medical imaging. As in medical imaging, military images are often difficult to interpret due to noise, spatial resolution limits, changing background conditions and uncertainty about the materials being imaged. We propose to vastly improve the results from hyperspectral imaging systems by modeling the physics that governs surface reflectivity; using the model to predict reflectivity of a few selected materials; validating the reflectivity model by acquiring spectral signatures of the materials with a characterized and calibrated hyperspectral imaging sensor; identify appropriate image- processing techniques to correct for instrument limitationsl; and analyzing the input and output signal-to-noise ratio of the sensor data and use this information to simulate a stochastic resonance-like (fixed threshold) approach to object identification. At the end of Phase I of this project we will provide a well defined approach and software to characterize the reflection properties of any material under test; a well defined approach and software to characterize imaging sensors of interest in hyperspectral imaging; image analysis procedures and software/hardware implementation to reduce the hyperspectral data set; and image segmentation/fusion procedures and software/hardware implementation to identify objects of interest.