针对图像去噪的速度以及可能出现的阶梯效应等问题进行了研究，提出了一种高效的图像去噪算法。该算法在贝叶斯框架下，首先引入调和模型作为原始图像的先验模型，并用伽马分布作为未知参数的先验分布模型；然后，用变分近似的方法推导最大后验概率；基于此推导过程，同步地估计原始图像和未知参数的最优值，实现图像去噪。实验结果证明了该算法的高效性，通过与其它算法的比较，该算法体现了速度快、效果好的优点，且去噪后的图像不会出现阶梯效应等问题。

An efficient algorithm was presented to solve the problems of the speed and the staircase effect in the image denoising. In the Bayesian framework, firstly, the harmonic model was introduced as the prior model of the original image, and the Gamma distribution was supposed to be the prior distribution model of the unknown parameters. Secondly, the maximum posteriori probability was deduced using the variational method, based on which the original image and the unknown parameters were estimated simultaneously to remove the noise in the observed image. The experimental results show the efficiency of the proposed algorithm. Furthermore, compared with other similar algorithms, the proposed algorithm shows the competitive performance on the speed without bringing the staircase effect to the denoised images.