基于数据驱动紧框架的含泊松噪声的图像恢复

摘要

This paper first proposed a data-driven tight frame based variational model for image restoration with Poisson noise.With this model,it acted weighted l2 norm term as fidelity term and l1 norm term as a regularization term contains a data-driven tight frame.Then,it also proposed reweighted split Bregman algorithm to solve this model.In addition,it also extends the proposed model and algorithm to remove mixed Poisson-Gaussian noise.Finally,it provided supporting numerical experiments and PSNR indicators to assess the model evaluation results.The results show that the algorithm is feasible and effective.%首先提出了基于数据驱动紧框架的含泊松噪声的图像恢复变分模型.在该模型中,赋权的l2范数项作为保真项,包含数据驱动紧框架的l1范数项作为正则项.然后,又提出了解该模型的重新赋权的分裂Bregman算法.另外,又将所提出的模型与算法拓展应用到了合泊松高斯混合噪声的图像恢复中.最后,利用仿真实验以及PSNR指标对该模型的图像恢复效果进行评估,评估结果表明该算法可行、有效.