针对显微图像盲复原算法存在的计算量大、振铃效应以及噪声敏感的问题，提出贝叶斯框架下两次引导滤波的快速盲复原算法。利用显微图像成像原理中基于深度信息估计点扩展函数的概率模型，构建了贝叶斯框架下盲复原的最小优化问题；通过分析最大后验概率的最小优化问题求解过程，推出了实施引导滤波器可快速求解优化问题的结论；为有效去除振铃和噪声，设计了两次引导滤波的求解方案，其将第一次引导滤波求解的结果作为优化问题的二次输入。实验结果表明，复原结果的像素误差率约为0.04，较常用盲复原算法的复原准确度提高了约20%，运行时间也大幅缩短，该方法能有效应用于显微视觉下微装配散焦图像盲复原的工程实践中。

To solve the problems of large computation cost, ringing and noise sensitivity in blind restoration algorithms for microscopic images, the blind restoration algorithm under Bayesian framework based on two guided filterings is proposed. The depth information of microscopic image is used to estimate the probabilistic model of point spread function, and a minimum optimization problem under the Bayesian framework is built. The guided filtering is applied to searching the optimal solution through analyzing the solving scheme of the minimum optimization problem of the maximum posterior probability. The solution scheme of the two guided filtering algorithms is designed for removing ringing and noise, which means the restoration result of the first guided filtering will serve as input of the optimization problem again. Experimental results show that the pixel error rate of recovery result is around 0.04, which increases by 20% compared to those of other commonly used algorithms, and the running time is significantly shortened. The proposed algorithm can be used in assembly of the micro-structures for defocused image blind restoration.