Noise-Robust Branch Feature Measure for Infrared Small Target Detection in Complex Scenes

Infrared (IR) small target detection is easily affected by background clutter, leading to a decrease in robustness. This letter proposes a noise-robust branch feature measure (BFM) for IR small target detection. The core idea is to ignore the background clutter in the neighborhood of small targets to improve the algorithm’s robustness. First, we establish an 8-branch model (8BM) in eight directions to achieve a robust reconstruction of the original target matrix. Second, a branch local contrast (BLC) operator is established for background suppression, which adaptively selects the minimum gray value as the branch background to ignore the influence of noise. At the same time, a branch local gradient (BLG) operator is established to enhance the target. Finally, the fusion of BLC and BLG can effectively suppress background clutter and improve the saliency of the targets. The area under the curve (AUC) value is better than 0.95, the SCRG value is better than 699.64, and the BSF value is better than 49.17 on four real image sequences. The experimental results show that the BFM method proposed in this letter has high robustness and good detection performance.