Neural network denoising in polygon meshes

One of the most important problems of creating 3D models with the aid of three-dimensional scanning systems is automatic processing to eliminate noise obtained due to the application of scanning devices with insufficient accuracy. The aim of the study is to develop a fully automatic approach for solving the problem of denoising in polygon meshes acquired after three-dimensional scanning. The principal method to overcome this is the application of neural networks that allow processing of polygon meshes to be performed automatically. The article presents an overview and comparative analysis of existing methods of denoising in polygon meshes. The mathematical formulation of noise elimination problem is provided. The description of the algorithms used to prepare data for neural network training is given. The method of polygon meshes filtering by the means of a bilateral filter, the method of principal components for reducing the dimension of data, the k-means clustering algorithm, the algorithm for updating vertex positions by updated face normals are employed. Details of a fully connected feedforward neural network implementation are described. The results of the study are outlined. The analysis of the findings is carried out utilizing the quality metrics of the Hausdorff distance and the average value of the angle between the normals of polygon meshes with and without noise.

1. Tomasi C., Manduchi R. Bilateral filtering for gray and color images. Computer Vision. Sixth International Conference on IEEE. 1998;1:839–846.

2. Fleishman S., Drori I., Cohen-or D. Bilateral mesh denoising. ACM Transactions on Graphics. 2003;22(3):950–953.

3. Eisemann E., Durand F. Flash photography enhancement via intrinsic relighting. ACM Transactions on Graphics. 2004;23(3):673.

4. Petschnigg G., Szeliski R., Agrawala M., Cohen M., Hoppe H., Toyama K. Digital photography with flash and no-flash image pairs. ACM Transactions on Graphics. 2004;23(3):664.

5. Zhang W., Deng B., Zhang J., Bouaziz S., Liu L. Guided mesh normal filtering. Computer Graphics Forum (Special Issue of Pacific Graphics 2015). 2015;34:23–34.

6. Sun X., Rosin P.L., Martin R.R., Langbein F.C. Fast and effective feature-preserving mesh denoising. IEEE Transactions on Visualization and Computer Graphics. 2007;13(5):925–938.

7. Wei M., Yu J., Pang W.M., Wang J., Qin J., Liu L., Heng P.A. Bi-normal filtering for mesh denoising. IEEE Transactions on Visualization and Computer Graphics. 2015;21(1):43–55.

8. He L., Schaefer S. Mesh denoising via l0 minimization. ACM Transactions on Graphics. 2013;32(4):64.

9. Yadav S.K., Reitebuch U., Polthier K. Mesh Denoising Based on Normal Voting Tensor and Binary Optimization. IEEE Transactions on Visualization and Computer Graphics. 2018;24(8):2366–2379.

10. Li X., Zhu L., Fu C.-W., Heng P.-A. Non-Local Low-Rank Normal Filtering for Mesh Denoising. Computer Graphics Forum. 2018;37(7):155–166.

11. Wang P.-S., Liu Y., Tong X. Mesh denoising via cascaded normal regression. ACM Transactions on Graphics. 2016;35(6):1–12.

12. Wang C., Cheng M., Sohel F., Bennamoun M., Li J. NormalNet: A voxel-based CNN for 3D object classification and retrieval. Neurocomputing. 2019;323:139–147.

13. Gantmaher F.R. Matrix theory. Moscow, Nauka; 1966. 576 p. (In Russ.)

14. The Great Russian Encyclopedia. Moscow, The Great Russian Encyclopedia; 2004–2017. (In Russ.)

15. Kingma D.P., Ba J. Adam: a method for stochastic optimization. ICLR. 2015;1:1–15.

16. Agarap A.F. Deep learning using rectified linear units (RELU). Arxiv. 2018;1:1–7.

17. Aspert N., Santa-Cruz D., and Ebrahimi T. MESH: Measuring Errors between Surfaces using the Hausdorff Distance. Proc. IEEE ICME. 2002;1:705–708.