Fast Low Rank and Sparse Decomposition Based on Greedy Bilateral Smoothing for Infrared Small Target Detection

The efficient and accurate detection of infrared small targets under various heterogeneous backgrounds has always been a key issue that needs to be addressed. To address this issue, this study presents a fast and robust low rank and sparse decomposition algorithm for infrared small target detection. Firstly, an infrared image patch model is constructed based on the local autocorrelation characteristics of the image, where the original infrared image can be transformed into a new infrared image through matrix decomposition and vector reconstruction. Leveraging the support of the low-rank approximation theory, the small target detection is formulated as an optimization problem involving a low-rank matrix and a sparse matrix. Specifically, in order to improve optimization efficiency, greedy bilateral smoothing is employed to model low rank backgrounds, resulting in a significant improvement in the efficiency of the detection algorithm. Then, a optimization algorithm framework based on alternating projection is designed to achieve an accurate separation of target and background. Finally, the adaptive threshold segmentation operation is adopted to extract the target. The experimental results on eight real infrared sequences demonstrate that the greedy bilateral smoothing exhibits powerful background suppression capability and can achieve higher signal-to-noise ratio gain, while maintaining stable detection performance, even in the presence of different additional noise interferences.