Answers are grounded in this paper and cite their sources.
Low-rank decomposition models excel in small target detection due to their strong background separation. However, most existing methods face real-time processing challenges due to the high cost of singular value decomposition (SVD) and additional regularization terms that increase computational complexity. Thus, fast and accurate infrared small target detection in complex backgrounds remains a critical challenge. To address this issue, we adopt the idea that an imprecise rough solution can retain structural information to regularize the final result. We use the efficient spectral residual method for initial background estimation and design it as a convex regularizer. With inherent low-rank properties, this regularizer replaces the nuclear norm operation in traditional low-rank decomposition, significantly reducing computational costs. Furthermore, we employ bimodal slice-based graph Laplacian regularization. Graphs are constructed with slices as vertices in both temporal and spatial dimensions. This approach more comprehensively captures local geometric features while avoiding the complexity of pixel-level graph construction. In addition, we combine well-designed single-frame priors and low-cost multi-frame motion posterior information to improve target extraction accuracy. Experimental results show that the proposed method outperforms state-of-the-art approaches in both detection performance and efficiency while demonstrating greater robustness in complex scenarios. The codes of the proposed method are available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/GXL257/Proposed</uri>.
No takes yet. Share an insight, caveat, or question.
Ge et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: