贝叶斯框架下的总变分图像去噪算法

针对经典去噪模型易造成图像细节丢失以及确定性算法无法自动估计去噪过程中的未知参数等问题，提出一种新的图像去噪算法.该算法在贝叶斯框架下，用总变分模型（TV）和伽马分布分别刻画原始图像及未知参数的统计特征，并基于最大联合分布的推导，估计最优原始图像.总变分模型使最终的能量泛函非线性且不可微分，因此，引入迭代重加权最小二乘法（IRLS），通过迭代的方式用加权的L2范数逼近L1范数来表示图像的统计模型.实验结果表明，该算法可有效去除图像的噪声，提升去噪速度，使所恢复的图像在实际视觉效果和信噪比等方面均优于其他同类算法.

Total variation based image denoising algorithm in Bayesian framework

Aiming at such problems as the image detail loss caused by classical denoising models and the unable estimation of unknown parameters in denoising process by deterministic algorithms, a new image denoising algorithm was proposed. The algorithm used the total variation model and Gamma distribution to depict the statistical characteristics of the original image and unknown parameters in Bayesian framework, respectively. The optimal original image was estimated through deriving the maximum joint distribution. Due to the nonlinearity and non differentiability of the resulting energy functional caused by the total variation model, the iteratively reweighted least squares method was introduced to represent the image statistical model by approximating the L1 norm with a weighted L2 norm through an iterative scheme. The experimental results demonstrate that the proposed algorithm can efficiently remove the image noise and improve the denoising speed. Compared with other similar algorithms, the proposed algorithm can give the denoised images with better visual effect and higher SNR value.