Infrared small-target detection via tensor construction and decomposition

Infrared small-target detection of a low signal-to-noise ratio (SNR) has been an important research hotspot. Small targets lack texture and detailed information; in addition, the background tends to be complex and diverse. These factors mean small targets can be easily submerged, resulting in a low probability of detection and a high false alarm rate. In the absence of texture and detailed information in infrared images, neighbouring spatial information is a good complement. In this paper, a novel method is proposed, which introduces three-order tensor construction and decomposition (TCD) to fully utilize neighbouring spatial information. Furthermore, discontinuous multi-scale windows are designed to achieve a more robust detection performance. Experiments are performed on three real infrared datasets and the results demonstrate the effectiveness of the proposed TCD.