Sparse representation based infrared small target detection via an online-learned double sparse background dictionary

Abstract

For infrared small target detection, traditional sparse representation based methods decompose infrared images either over a background dictionary constructed using original image data or over a target dictionary generated by Gaussian intensity model. However, these methods can hardly be employed in practice, due to the time-consuming decomposition and the interference from heavy background clutter. To achieve a more efficient detection of infrared small targets, in this work, we propose a novel method based on sparse representation via an online-learned double sparse background dictionary. First, we employ the double sparsity model to construct a background dictionary and further propose a modified online learning algorithm to train the dictionary. Second, we use the sparse representation model to decompose the target image into background component, target component and noise component. Third, we propose an edge clutter suppression strategy to improve the robustness of detection. Experimental results on five real sequences show that the proposed method can not only detect targets precisely with low false alarm rate, but also suppress background clutters effectively.