Directional segmented matched filter for hyperspectral images

We propose a new type of a segmented matched filter, called a directional segmented matched filter (DSMF), based on geometric considerations between the target, background, and a given pixel. We use a modified k-means segmentation algorithm that has produced the background suitable for our detectors. We demonstrate numerical results on the performance of two versions of DSMF on two different targets. The two versions perform very differently, apparently depending on the type of target and the background structure.