摘要
研究红外焦平面的非均匀校正对监控系统和军事有着特殊的意义。针对传统神经网络法非均匀校正算法存在收敛速度慢和不稳定的缺点,提出了一种新的基于场景的IRFPA非均匀性校正算法。该算法先将焦平面上的各像素点值和他周围的8个像素点值做一次排序,选择排在中间的5个像素值求平均作为该点的新像素值。再利用一种改进的神经网络法对红外图像再做一次非均匀校正。实验结果表明,新算法的非均匀校正效果比原来的神经网络算法和均值滤波算法都有明显的提高。还引用了一种新的收敛因子的估算方法,计算结果得出该方法能较准确地估算出收敛因子在自适应迭代公式中收敛时的范围,提高了校正算法的收敛速度。
It has a special meaning to study non-uniformity correction algorithm for infrared focal plane array （IRFPA）. In order to improve the convergence speed and non-stability in traditional neural network non-uniformity correction algorithm, the article was designed a new scene-based non-uniformity correction algorithm in IFPA. Firstly, The algorithm was that the pixel and its around eight pixels in IFPA got order from small one to big one and choose five Pixels that are close to the mid one in nine ordered pixels to obtain their mean as a new pixel. Finally we use an improved neural network algorithm to do a non-uniformity correction to infrared image again. We could get such result： the new algorithm have better correction effect than the improved neural networks algorithm. Besides, we try to use a new estimating algorithm to calculate precisely the scope of the recursive constant in equations, which could get faster convergence speed.
出处
《红外技术》
CSCD
北大核心
2008年第9期520-523,共4页
Infrared Technology
基金
厦门大学"985"二期"信息科技创新平台"0000-X07204
关键词
红外焦平面
非均匀校正
排序滤波
收敛因子
infrared focal plane array
non-uniformity correction： ordering filter
convergence constant
