As is well-known that the total generalized variation model performs well in reducing the staircasing effect while removing noise, but it tends to cause the undesirable edge details blurring. To overcome this drawback, the current paper introduces the non-convex restriction into the total generalized variation regularizer and constructs an improved edge-preserving optimization model for Poissonian images restoration. For solving the minimization problem, we propose an efficient alternating minimization method by skillfully combining the classical iteratively reweighted \(\ell _1\) algorithm and primal-dual framework. Some visual experiments presented in the illustration section, which are compared with some related denoising methods, demonstrate the better performance of the developed scheme in staircase artifacts reduction and image features protection. Besides, the measurable comparisons also indicate that our outcomes enjoy the best restoration accuracy against other popular competitors.

众所周知，全广义变异模型在去除噪声的同时减少阶梯效应的效果很好，但容易造成不希望的边缘细节模糊。为了克服这个缺点，本文将非凸限制引入到总广义变异正则化器中，并构建了一种改进的泊松图像恢复边缘保留优化模型。为了解决最小化问题，我们通过巧妙地结合经典的迭代重加权\(\ell _1\)算法和原始对偶框架。插图部分中展示的一些视觉实验与一些相关的去噪方法进行了比较，证明了所开发的方案在减少楼梯伪影和图像特征保护方面的更好性能。此外，可测量的比较还表明，我们的结果与其他流行的竞争对手相比具有最佳的恢复精度。