Abstract: The locally statistical characters of the random valued impulse noise image are studied in detail. And the gray-levels occurrence probability is defined to measure the noise reliability of the image pixels. Then a nonlinear algorithm called minimum occurrence probability median (MOPM) filter is presented to remove the random valued impulse noise from grayscale images. The presented algorithm performs noise detection and filtering at multiple resolutions based on the biological concept of foveal vision. Computer simulation results show that the presented algorithm outperforms the other filtering algorithms in the aspects of filtering ability, adaptability and details preserving ability, which is very close to the ideal median filter, especially for highly corrupted images.

Key words: random valued impulse noise, gray-levels occurrence probability, noise reliability

摘要： 本文详细分析了随机值脉冲噪声污染图像的局部灰度统计特征，定义一种噪声可信度的估计函数—灰度等级共现几率，并在此基础上提出了共现几率极小的中值滤波算法。算法模拟生物视觉在注视过程中的变分辨率特性，以不同分辨率对噪声进行多层次的定位及滤波。实验结果表明，本文算法可以有效地滤除噪声，并较好地保留图像的边缘细节信息，其滤波效果比其它算法更接近理想的中值滤波，尤其是对于噪声高度污染的情况。

关键词: 随机值脉冲噪声, 灰度等级共现几率, 噪声可信度