Effective and robust infrared small target detection based on directional improved structure tensor

Abstract

Small target detection under complicated background is a research hotspot and challenging problem. To increase the ability of small target detection under complicated background, a simple but effective target detection approach based on directional improved structure tensor is designed. Firstly, considering that small target region will be omni-directional while background edge clutters will tend to expand in some specific directions, the directional structure tensor (DST) is constructed by incorporating eight-direction filter template. Then, the directional positive small eigenvalue of the structure tensor (DPSTSE) is computed to effectively perceive and locate the junctions of small target and surrounding background in all directions. Finally, the DPSTSE maps in all directions are integrated together to distinguish small targets from strong background clutter. Experiments illustrate that the presented method is better than some advanced algorithms in clutter suppression ability and target detection effect. Moreover, the presented method has high computational efficiency and potential for real-time processing.