基于变分正则化的乘性噪声图像去噪综述

Based on Variational Regularization of the Multiplicative Noise Image Denoising Review

为研究变分正则化乘性噪声去除的发展及现状,本文首先根据保真项的不同特征将模型分为方差–均值、最大后验概率、I-散度及混合保真项乘性噪声去噪模型。其次分别对此进行了详细介绍并从保真项及正则项的角度出发对模型进行了总结。接着通过实验对不同类型的保真项模型的效果进行直观展示,并给出恢复图像的峰值信噪比值及实验运行时间。最后提出了自适应参数的设定、求解速度的改善、图像的分解去噪、特定图像去噪等几个方面关于乘性噪声去除未来发展的趋向和展望。

In order to study the development and current situation of variational regularization multiplicative noise removal, this paper first divides the models into variance mean, maximum a posteriori esti-mate, I-divergence, and hybrid fidelity multiplicative noise removal models according to the differ-ent characteristics of fidelity terms. Then it introduces them in detail and summarizes the models from the perspective of fidelity terms and regularization terms. Then through experiments the ef-fect of model fidelity term for different types of visual presentation, and restore the image peak signal-to-noise ratio value and test running time. Finally, the trend and prospect of multiplicative noise removal in the future are put forward, such as the setting of adaptive parameters, the im-provement of solution speed, image decomposition denoising and specific image denoising.