对红外焦平面阵列成像系统而言, 基于场景的非均匀校正技术是处理固定图案噪声的关键技术。现有的非均匀校正算法主要被收敛速度和鬼像问题所限制。提出一种新的基于恒定统计算法的自适应场景非均匀校正技术。利用红外图像序列的时域统计信息结合提出的α修正均值滤波来估计探测器的参数, 通过减少样本的渐进方差估计, 完成成像系统的非均匀性校正。通过模拟和真实的非均匀性图像对算法的性能进行评价。实验结果表明, 在继承恒定统计算法快速收敛的同时, 图像峰值信噪比较恒定校正法及常系数α校正算法分别有445%和329%的提升, 图像鬼像问题有明显改善。

For infrared focal-plane array imaging system, scene-based non-uniformity correction is key technique to deal with fixed pattern noise. Existing algorithms are mainly restricted by convergence speed and ghosting artifacts. In this paper, a novel adaptive scene-based non-uniformity correction technique is presented, which is based on constant-statistics method (CS). Utilizing temporal statistics of infrared image sequences, the proposed method applies an alpha-trimmed mean filter to estimate detector parameters and minimize sample asymptotic variance estimate. Performance of proposed technique is evaluated by simulation and real non-uniformity image. Experimental results show the proposed method inherits characteristics of fast convergence of CS method and increases peak signal to noise ratio by 44.5% and 32.9% respectively, and image ghost problem is improved obviously.