Abstract: Aiming at estimating the unknown shape parameter p of generalized Gaussian noise in image, this paper proposes a novel approximated moment estimation algorithm of p. It extracts noise samples from the original image, and fits the Ratio Function(RF) of p by using piecewise function, yielding an approximated expression of p. Experimental results demonstrate that, when the noise samples are precise, the estimates are accurate to the second decimal place, especially in the case of noise with uniform distribution, the estimation error is 0.3 smaller than that of the traditional algorithm. As for the imprecise noise samples, the estimation precision is dependent on the threshold parameter K.

Key words: generalized Gaussian noise, shape parameter, moments estimation, numerical fitting

摘要： 针对图像中广义高斯噪声的形状参数p通常为未知的问题，提出一种用于计算p的近似矩估计算法。从原图像中提取噪声样本，采用分段函数对p的比率函数进行数值拟合，从而得到p的近似表达式。实验结果表明，当噪声样本值准确时，p的估计值能精确到小数点后2位，尤其适用于均匀噪声的情况，估计误差比传统算法小0.3；当噪声样本值不准确时，估计精度与门限K的选择有关。

关键词: 广义高斯噪声, 形状参数, 矩估计, 数值拟合

CLC Number: 

• N945