A neighborhood model for detection in hyperspectral images

The neighborhood model provides a moderate complexity method of introducing the concept of smoothness into a detection problem. As tested here, the smoothness is reduced to a simple scalar quantity whose probability is easily computed. The concept is fairly general, moving from vector matched filter processing as originally formulated to any scalar image. The result is a nonlinear filter which is edge preserving and classifier-enhancing, resulting in improvements in the ROC curve in all classifiers tested, the neighborhood modeling.