This paper focuses on the denoising and enhancing of 3D data. Our aim is to denoise the 3D blocks without blurring relevant details such as edges. In order to achieve this task, we propose a new 3D tensor diffusion process based on the classical Coherence Enhancement Diffusion (CED) proposed by Weickert. The framework relies on the computation of a tensor providing information on the local orientation of structures. Then, the process is steered according to this directional information. In addition, the smoothing effect is adjusted according to a confidence measure that takes into account the regularity of the local structure. Through applications samples we will show the efficiency of our method on synthesized 3D data.

Keywords: 3-D filtering, Anisotropic diffusion, Confidence measure, Structure tensor.