Total Variation for Image Denoising Based on a Novel Smart Edge Detector: An Application to Medical Images

In medical imaging applications, diagnosis relies essentially on good quality images. Edges play a crucial role in identifying features useful to reach accurate conclusions. However, noise can compromise this task as it degrades image information by altering important features and adding new artifacts rendering images non-diagnosable. In this paper, we propose a novel denoising technique based on the total variation method with an emphasis on edge preservation. Image denoising techniques such as the Rudin–Osher–Fatemi model which are guided by gradient regularizer are generally accompanied with staircasing effect and loss of details. To overcome these issues, our technique incorporates in the model functional, a novel edge detector derived from fuzzy complement, non-local mean filter and structure tensor. This procedure offers more control over the regularization, allowing more denoising in smooth regions and less denoising when processing edge regions. Experimental results on synthetic images demonstrate the ability of the proposed edge detector to determine edges with high accuracy. Furthermore, denoising experiments conducted on CT scan images and comparison with other denoising methods show the outperformance of the proposed denoising method.