Robust small infrared target detection using multi-scale contrast fuzzy discriminant segmentation

Infrared images with complex cloudy-sky and sea-sky backgrounds have low contrast and a low signal-to-clutter ratio (SCR), which makes it extremely challenging for conventional small target detection methods to accurately detect small targets. This study proposes a novel method for detecting small infrared targets by using spectral, spatial, and temporal information to improve the detection performance. First, to achieve background suppression and target enhancement, the proposed approach implements a parallel reconstruction with the spectral information to generate the fusion saliency map. Then, multi-scale contrast fuzzy discriminant segmentation (MCFDS) is used to measure the certainty of the target boundary using the spatial information. The detection task is regarded as a fuzzy discriminant segmentation problem, which can determine the target boundary accurately. Subsequently, position prediction with optical flow estimation and forward pipeline filtering are adopted to achieve multi-frame target recognition and reduce the false alarm rate effectively. The proposed algorithm is evaluated on real infrared thermal image sequences of small targets against complex cloudy-sky and sea-sky backgrounds. The results show that it can achieve stable, continuous, and robust detection for different target sizes and SCR values as well as for multiple targets and/or background types compared with conventional baseline methods, such as top-hat, ILCM, PQFT, MPCM and TM-SCR. In addition, mathematical proofs are provided to elucidate the proposed approach.