非局部TV正则化的图像泊松去噪模型与算法

Image Poisson Denoising Model and Algorithm Based on Nonlocal TV Regularization

针对图像泊松噪声去除问题,提出一种基于非局部全变差正则化的图像去噪模型。在Bayesian-MAP框架下,采用负自然对数泊松似然函数作为保真项。结合图像的非局部相似性与梯度模稀疏性先验,构造非局部TV正则项,建立图像泊松去噪非局部TV正则化模型。利用变量分裂法和交替最小化方法对模型进行求解。实验结果表明,所提模型和算法能够较好的处理图像泊松去噪问题。与其它图像泊松去噪模型和算法比较,模型的图像恢复性能无论是在视觉效果还是在客观评价指标上都有明显改善。

Concerning the removal problem of image Poisson noise, a novel image denoising model was proposed based on nonlocal total variation （TV） regularization. In the framework of Bayesian-MAP estimation, the negative-log Poisson likelihood function as the fidelity term was utilized. Combining the nonlocal similarity prior and gradient sparsity priori of the natural image, nonlocal TV regularization term was constructed to build the image Poisson denoising optimization model. The optimization model could be reduced to a series of convex minimization problems that could be efficiently solved with a combination of the variable splitting method and the alternating minimization method, leading to fast and easy-to-code algorithms. Experimental results show that the proposed method is effective , and it outperforms than other methods both on objective criterion and visual fidelity.