메쉬 디노이징을 위한 결합형 양방향 노멀 필터링

Normal filtering is an effective mesh denoising technique where perturbed normals are first rectified followed by vertex adjustment on a new normal field. However, in most filters, important features such as edges, corners or small-scale details are not preserved, or shape distortions inevitably occur during normal refinement. This is due to the characteristic that a filter generally takes measurement signal as a centroid of the range filter. In this paper, we propose a joint bilateral filter that uses the reference data obtained by analyzing the measurement data itself. The algorithm classifies every face into feature or non-feature face, and normal filtering is then performed on non-feature and feature face separately. The proposed normal filter outperforms the other methods for highly corrupted mesh surface, even for impulse noise.