图像正则化扩散行为及统一处理框架

Diffusion behavior of regularization and a unified framework

PDE的图像正则化是一种基于扩散思想的非线性滤波方法,是解决降噪、伪影去除、结构增强等底层视觉问题的最有效方法之一,目前针对此类算法的统一分析框架还较为少见。基于3种典型PDE正则化算法的扩散行为,提出了一种基于扩散张量的图像正则化算法分析框架,对于此类算法的分析、开发和拓展具有重要意义,最后通过实验验证了框架的有效性。

PDE-based image regularization is a nonlinear filter based on the diffusion principle,and also is one of the most efficient solutions for low-level vision consisting of denoising,artifact elimination and structure enhancement. Nevertheless,the unified frameworks for analyzing the PDE-based regularizations are relatively rare. This paper reviews three typical PDE-based methods by analyzing their diffusion behavior,then,proposes a unified analysis framework based on diffusion tensor. It exhibits the fundamental significance for analysis and development,also extends the PDE-based regularization methods. The feasibility of our proposed framework is verified via the experiments.