Texture Image Denoising Algorithm Based on Structure Tensor and Total Variation

Abstract

For the existing problems of staircase effect, edge blur and uncertainty of parameter selection in the process of image denoising and recovery of variational partial differential equations, a novel total variation restoration model based on image structure tensor(STTV) is proposed. We introduce image structure tensor to construct the image structure control function instead of using Lagrange multiplier and local structure information to control Diffusion process, which has the performance of adjusting the balance of regular item and fidelity item in TV model according to different local structure information and keeping better detail features. Theoretical analysis and experiment comparing with other methods illustrate that STTV model is able to describe the image edges, textures and smooth areas more accurately and subtly, which has overcome staircase and over-smoothing effects brought by other TV models and removed the noise while preserving significant image details and important characteristics. the value of peak signal to noise ratio(PSNR) is also improved.