一种改进的各向异性SUSAN 红外弱小目标检测方法

摘要

When anisotropic SUSAN filtering is used to detect the infrared dim target , the edge’ s o-rientation obtained only with four pixels is quite different from the real edge ’s orientation , which causes the filtered image to preserve numerous rougha-nd-tumble infrared backgrounds and is disadvantageous to detect the dim target.Therefore, a method to optimize the edge ’ s orientation is given, which combines coordinate transformation so as to make full use of the pixels in part image to estimate the edge ’s orienta -tion, and the optimized edge ’ s orientation which is most accurate to the real edge ’ s orientation would be given adaptive in the use of LOG operator .Then the average gray difference between the part image and the pixel which is in the middle of the part image is certain as the threshold of the anisotropic SUSAN fil-tering .The experiments show that this algorithms has the better ability of restraining the infrared back -ground , highlighting the target, effectively improving the GSCR of the infrared image , as well as preferably avoiding the emergence of the illusive target , thus, it improves the ability to detect the infrared dim target in a single frame.%运用各向异性 SUSAN滤波进行红外弱小目标检测时，考虑到仅依靠4个像素点所确定的边缘方向往往与实际边缘方向存在较大误差，导致滤波后的图像依然保留着较多的背景杂波而不利于弱小目标的检测，因此提出一种边缘方向优化方法，结合坐标变换尽可能多地利用局部图像的像素点来参与边缘方向的估值，并运用高斯-拉普拉斯算子自适应地选择最优化的边缘方向角，使其更贴近于实际边缘；然后将局部图像与中心像素的灰度差均值确定为各向异性SUSAN滤波算子的阈值。实验表明：本文算法具有更好的背景抑制能力，突出目标，较好地提升了红外目标信噪比增益，并有效避免虚假目标的产生，提高了单帧红外弱小目标的检测能力。