Robust and fast infrared small target detection based on pareto frontier optimization

Infrared small target detection is a challenging task due to the small target size and complex background clusters. The matrix factorization-based algorithms can work well in addressing this issue. In this paper, to address the residual over-optimized issue in matrix factorization, a novel Pareto frontier optimization-based algorithm is proposed to obtain the best trade-off between a given noise level and the optimal objective value. To suppress background outliers, a robust Huber penalty and non-negative constraints are introduced in matrix factorization. To solve the nuclear norm regularization faster, a randomized SVD algorithm is adopted. Finally, the proposed algorithm is compared with 11 representative detectors on 6 proposed test sequences, which have more than 1500 frames. The experimental results demonstrate the effectiveness of the proposed algorithm in improving speed and suppressing outliers. The source codes and data are available on https://github.com/wahahamyt/ISTD.