Infrared small target detection based on multi-directionality and sparse low-rank recovery

Abstract

Infrared small target detection plays an important role in diverse fields, while complex backgrounds make it hard to identify targets accurately. In this paper, we propose a novel method to handle the impact of complex backgrounds. First, we analyze the infrared small target images and model this task as a constrained optimization problem by utilizing the sparsity and multi-directional intensity spread of target images, the self-similarity of background images. By modeling the characteristics of images, mathematical constraints are proposed to enhance targets. Second, we propose Multi-directionality and Sparse Low-rank Recovery (MDLR) method. On the basis of Principal Component Analysis, the algorithm considers characteristics of infrared small target images to detect targets accurately. Finally, we solve the proposed optimization objective function by alternating direction method of multipliers. Comprehensive evaluations on real data sets show that MDLR achieves a balance detection performance of target enhancement, background suppression and computational efficiency in complex backgrounds.