自适应的非局部总变分图像复原算法

Adaptive Non-local Total Variation for Image Restoration

边缘和纹理信息是图像的重要特征信息,在复原过程应得到有效地保持.为此提出了基于非局部总变分的自适应复原模型.新模型的能量泛函包含两项:自适应正则项和数据保真项.自适应正则项的设计是建立在非局部正则项的基础上,主要用其保持纹理;另一方面正则项融入了边缘探测函数,使得模型可以根据局部边缘信息自适应地实施不同程度的扩散力,即在同质区域作用一个较大的扩散过程,避免出现分片常值;在边缘附近作用一个相对平和的力,使得图像边缘得到有效地保持.本文模型利用Gateaux导数计算其扩散方程,通过有限差分进行数值模拟.实验结果表明,所提模型在视觉效果和运行效率方面取得了较好的结果,并且在边缘锐化和纹理保持方面,本文模型相对于经典的TV模型和NLTV模型具有较大的优势.

Edge and texture information are important for the given image,which should be preserved in image processing and analysis.In order to preserve this information,we propose a novel adaptive image restoration model based nonlocal total variation.The proposed model consists of two terms：adaptive regularization term and image fidelity term.The adaptive regularization term is designed based on the non-local operator,which can deal with texture and fine information well.On the other hand,the proposed model performs adaptively the different forces on the observed image according to the local edge information in the image and avoids blurring edges based on edge indicator function.The diffusion equation is obtained by Gateaux derivative,and be computed by the finite difference method.This adaptive approach is applied to restore the noisy blurred images.Experimental results show that the proposed model obtain good results in terms of visual perception and operational efficiency.In addition,the proposed model,compared with classical TV model and NLTV mode,can achieve better results than other models in terms of edge and texture information.