红外弱小目标检测是图像处理的难点之一, 许多研究人员提出了不少检测方法。针对复杂背景与强杂波干扰下图像信杂比(Signal-to-Clutter Ratio, SCR)低造成的目前检测方法易受伪目标干扰、虚警率高的问题, 提出了一种多信息融合的红外弱小目标检测算法。首先, 构建八向局部灰度残差信息图; 其次, 设计一个滑动窗口遍历整个图像, 将图像分为一系列局部图像块, 对局部图像块的强度均值进行约束, 获得局部强度均值约束信息图; 然后, 将局部图像块进一步划分为 12个方向块, 对每个方向块中像素的梯度方向进行约束, 获取梯度方向约束信息图; 最后, 上述 3个信息图像通过点积运算得到最终显著图, 并利用阈值分割实现弱小目标的分离。将该算法与 3种其它不同算法从信杂比增益(Signal-to-Clutter Ratio Gain, SCRG)、背景抑制因子(Background Suppression Factor, BSF)以及检测率与虚警率的接受者操作特征(Receiver Operating Characteristic, ROC)曲线方面进行对比。实验结果表明: 该算法具有更高的 SCRG、BSF和 ROC曲线下面积(Area Under the Curve, AUC), 不仅能有效地抑制背景杂波、剔除伪目标, 而且能准确地检测出红外弱小目标, 具有较高的检测率。

Infrared dim target detection is a challenging task in image processing, and several researchers have proposed various methods to address this challenge. However, under conditions of complex background and strong clutter interference, because of low signal-to-clutter ratio (SCR), current detection methods are susceptible to reaching false targets and having a high false alarm rate. To resolve this problem, an infrared dim target detection algorithm based on multi-information fusion has been proposed. First, the gray residuals map was obtained by calculating the 8-directional local gray residual. Second, the image was divided into a series of local image patches by using a sliding window, followed by the intensity mean of local image patches being constrained to achieve the local intensity mean constrained map. Lastly, the local image patches were further divided into 12 directional blocks; from this, the gradient direction constrained saliency map could be obtained by constraining the gradient direction of the pixels in each directional block. The final saliency map was subsequently obtained from the aforementioned three feature maps by dot product operation, and dim targets were separated by threshold segmentation. The proposed method was compared with LCM, MPCM, and LIG algorithm based on SCRG, BSF, and ROC curves. Experimental results demonstrate that the proposed method is superior than the other above-mentioned algorithms; it can strongly suppress background clutter and eliminate false targets, therefore, it is a valuable algorithm with low false alarm rate and a high detection rate.