Fig. 1. Multi-layer Riemannian graph. (a) Point cloud; (b) the first layer of Riemannian graph; (c) the second layer of Riemannian graph; (d) the third layer of Riemannian graph

Fig. 2. Normal propagation of sample points. (a) Ununified sample normal; (b) normal propagation result of the first layer; (c) normal propagation result of the second layer; (d) normal propagation result of the third layer; (e) sample normal after propagation

Fig. 3. Normal propagation error of top-layer Riemannian graph

Fig. 4. Point-cloud models for testing. (a) Hood; (b) fandisk; (c) Venus; (d) Anya; (e) Maitreya; (f) dragon

Fig. 5. Comparison of normal unification results based on single-layer and multi-layer Riemannian graph methods. (a) Single-layer Riemannian graph method; (b) multi-layer Riemannian graph method

Fig. 6. Comparison of normal unification results of fandisk model with different methods. (a) Method in Ref.[14]; (b) method in Ref. [12]; (c) proposed method

Fig. 7. Comparison of normal unification results of dragon model with different methods. (a) Method in Ref. [14]; (b) method in Ref. [12]; (c) proposed method

Fig. 8. Comparison of Poisson surface reconstruction of Maitreya model. (a) Method in Ref. [14]; (b) method in Ref. [12]; (c) proposed method