Image Factorization for Small Object Detection

Blind source separation techniques, specifically independent components analysis and Nonnegative Image Factorization have seen increasing use in the hyperspectral community for automated image exploitation. These techniques differ from more traditional image reduction methods such as principal components in that they make different statistical assumptions as to the nature of the image. As such, these techniques provide the potential for the development of exploitation techniques that better preserve spectral information associated with small targets that tends to be lost with more traditional statistical processing.