Total variational denoising using improved Adaptive Fidelity term

Denoising is an important part of digital image processing. Adaptive Fidelity term Total Variation (AFTV) method can effectively remove the noise and can preserve the edge and detail information of the image. However, noise variance should be given in the method, and the mothed is sensitive to noise, which will lead to unsatisfactory denoising results in edges of image. Therefore, an improved Adaptive Fidelity Total Variation algorithm is proposed. The method first uses the local variance of the noise image to initially estimate the confidence parameters, then the optimized parameters are obtained with anisotropic convolution. The experiments with several images demonstrate that the proposed method is superior to AFTV method at different noise levels.