时域高通滤波非均匀性校正是一种典型的基于场景的红外焦平面阵列非均匀性校正算法，但其易产生“鬼影”现象，影响校正效果.本文在时域高通滤波校正算法的原理基础上，分析了其校正过程中“鬼影”现象产生的原因，即由于全部图像信息的叠加而导致静止场景被滤除且运动场景会在当前位置留下反转的图像，从而形成“鬼影”.引入非局部均值滤波方法，提出了一种去“鬼影”的非局部均值滤波-时域高通滤波非均匀性校正方法.该方法首先采用非局部均值滤波将图像信息分离成高低频两部分（其中高频成分含有大部分噪音及非均匀性），并使用高频成分进行时域高通滤波算法中低通输出的递归运算，使得低通滤波后的图像含有较少的场景信息，从而可使校正输出图像含有较少的“鬼影”现象.采用两组真实红外序列图像进行验证，结果表明该算法不仅能获得较好的非均匀性校正效果，而且能较好地抑制时域高通滤波算法中的“鬼影”现象.

The nonuniformity correction algorithm with the temporal high-pass filter for infrared focal plane array is a typical scene-based nonuniformity correction technique, but it has the ghosting artifacts problem. Based on the basic principle of temporal high-pass filter algorithm, the superposition of the whole image information will result in the elimination of stationary scene and the stay of a reversed image which can be referred to as the ghosting artifacts on the current position of the moving scene. A new scene-based nonuniformity correction technique based on the non-local means filter was proposed to eliminate the ghosting artifacts in the temporal high-pass filter algorithm. In this algorithm, the original input image was separated into two parts, which were the high spatial-frequency part and the low spatial-frequency part. And then the output of the low-pass filter was estimated recursively only by using the high spatial-frequency part of the image which contains most of the noise and nonuniformity. The performance of this presented algorithm was tested with two infrared image sequences, and the experimental results show that the algorithm can significantly reduce the ghosting artifacts in THF and achieve a better nonuniformity correction effect.