Fisher信息在噪声估计精度分析中的应用

Application of Fisher Information in Noise Estimation Accuracy Analysis

采用Fisher信息以及相关的渐近正态性,分析基于极大似然方程估计的广义噪声模型的参数精度.理论分析结果表明,对于标准像素图像,用极大似然方程估计得到的加性噪声的参数误差大于信号相关噪声参数的误差;而对于归一化后的图像,参数的精度结果刚好相反.实验证明了理论分析的正确性.

We analysed the parametric accuracy of the generalized noise model estimated based on the maximum likelihood equation by using Fisher information and the associated asymptotic normality.The theoretical analysis results show that for standard pixel images,the parameter error of the additive noise estimated by using the maximum likelihood equation is larger than the error of signal-dependent noise parameter,while for the normalized images,the accuracy of the parameters is exactly the opposite.The experiments prove the correctness of the theoretical analysis.