Infrared small target detection based on local significance and multiscale

Infrared small target detection is a crucial component of infrared search and track systems. Existing methods based on the human visual system often reduce the saliency of targets, failing to effectively remove the background while preserving the scale information of the target. To address these issues, this paper proposes a novel filtering method based on local saliency and multi-scale analysis. Initially, we designed a new local saliency computation method that retains local contrast and standard deviation information, thereby enhancing the grayscale response of the target. Subsequently, we refined the calculation method for local multi-scale features of the target, considering the standard deviation of contrast differences as a significant factor influencing the target's scale. This preserves the scale characteristic information of the target. Finally, we employed an adaptive threshold segmentation fusion approach, combining the advantages of local saliency and multi-scale analysis while suppressing their deficiencies, to detect the final target. Through qualitative and quantitative experimental analyses, the proposed method demonstrates superior performance in suppressing background clutter and preserving scale feature information of small targets in infrared images. On public datasets, our method outperforms baseline approaches.