实时中子照相系统图像斑状噪声迭代滤波算法

Iterative filtering of speckle noises in real-time neutron radiography image

根据实时中子辐照图像“斑状”噪声的成因和特点,提出一种有效的迭代滤波算法。设计十字形噪声检测窗口,通过计算邻域一致性测度(NHM),将像素分为噪声和信号。对噪声进行迭代滤波,而对信号则不做任何处理。滤波是一个中值计算过程,同时窗宽可自适应调整。这种方案不仅避免了噪声在邻域的传播,且有较高的计算效率。实验结果表明,对于峰值信噪比(PSNR)为24.51dB的噪声污染图像,3×3中值滤波后PSNR只有26.81dB,而本文算法能将其提高到31.54dB,同时图像的视觉效果更好。

Real-time neutron radiography image is usually corrupted by speckle noise. An effective iterative filtering algorithm is presented based on the mechanism and characteristics of speckle noise. It consists of noise detecting and iterative filtering. A crisscross detection window is designed and each pixel is classified to be signal or noise via calculating the neighborhood homogeneous measurement （NHM）. The noise pixels are iteratively filtered and other valid pixels are preserved without modification The filtering is a process of median calculation and the size of window can be modulated adaptively. This scheme prevents the propagation of noises in the neighborhood and features highly computational efficiency. For the corrupted simulated image whose peak signal to noise ratio（PSNR） is 24.51dB, the results show that the presented approach can outperform 3 × 3 median filtering both visually and in terms of the PSNR values, which is improved from 26.81dB to 31.54dB.