Sector-based diffusion filtering

In this paper, we propose a new approach devoted to the denoising and the enhancing of strongly oriented 3-D images. In particular, the paper focuses on seismic data composed of a stack of layers disturbed by noise and broken by faults. The denoising of those data is a preprocessing used to improve the detection of the faults. Our method is based on an anisotropic forward and backward diffusion scheme, which takes advantage of the computation of a "regional" orientation. This approach allows the recovering of the plan, which is tangent to the current layer and the corresponding normal direction. Then the diffusion goes forward along the layer in order to smooth the noise, and backward along the normal to separate the layers.