Multi-scale Similarity Enhanced Guided Normal Filtering

In this paper, we propose a novel mesh denoising scheme in which multi-scale similarity is exploited to improve the performance of non-local normal filtering for feature-preserved mesh restoration. In our scheme, K-ring patches are used to identify multi-scale local structures around faces, and we compare the similarity between patches on multiple levels. The multi-scale similarities are subsequently computed by weighted similarity of patches. Finally, the center faces of similar patches are weighted by similarities in face normal filtering. Experimental results on different models indicate that the proposed method outperforms other local and non-scale-aware similarity based schemes in terms of both objective and subjective evaluations.