A New Algorithm for Local Background Suppression in Hyperspectral Target Detection

This paper deals with target detection in hyperspectral images based on local background suppression. Global approaches to background subspace estimation and suppression may be ineffective for target detection purposes. In fact, they tend to overestimate the background interference affecting a specific target. This typically results in a low target residual energy after background suppression, which is detrimental to detection performance. In this work, a local methodology is investigated that estimates the local background subspace over a local neighborhood of each pixel. By acting on a per-pixel basis, the proposed method adaptively tailors the estimated basis to the local complexity of background and it is expected to yield a higher target residual energy after suppression, thus benefiting to detection performance. Real hyperspectral imagery is employed to show the detection performance improvement offered by this approach with respect to a conventional global methodology.