An Adaptive Fidelity Term for Total Variation Denoising Method Using Local Structure Information

A new adaptive fidelity term for total variation denoising method is presented. Two requirements of local structure descriptors are discussed first, and then the defects of the linear structure tensor in local structural information extraction are analyzed. To make the total variation denoising method more capable of denoising, an adaptive fidelity term is constructed by using nonlinear structure tensor. Experimental results show the superiority of the new method in preserving geometric structure, such as edges, corners and so forth.