Infrared dim small target detection in complex scenes using an orientation-aware multiscale contrast measure
Abstract
Robust detection of infrared dim small targets is critical in long-range detection and precision guidance systems. Targets in complex backgrounds usually occupy only a few pixels with extremely low contrast, and are often buried in structured clutter such as clouds, sea-sky lines and strong noise, rendering single-frame detection challenging. To address this issue, we propose an orientation-aware multi-scale contrast (OMSC) measure. The method first constructs a set of linear weighting masks covering eight directions to exploit the prior that background clutter tends to extend along particular orientations, thereby achieving orientation-selective suppression. A detail-enhancement preprocessing is then applied to emphasize the target while flattening the smooth background. On this basis, a multiscale saliency contrast measure is defined and the directional responses are fused by a pointwise minimum rule to produce the final saliency map, from which the target is extracted by an adaptive threshold. Experiments on ten real infrared image sequences demonstrate that the proposed method outperforms several existing methods in both signal-to-clutter ratio gain and background suppression, and achieves a higher detection rate at a lower false alarm rate. The directional operations are mutually independent and suitable for parallel implementation.