RECOGNITION OF 3D OBJECTS BASED ON SPECTRAL INVARIANTS USING DEEP MACHINE LEARNING
Abstract
The paper proposes a method for recognition of three-dimensional objects using deep ma-chine learning. In computing systems, objects are often represented by 3D models as a set of poly-gons or surfaces describing a geometric shape. Search for relevant objects by recognizing on the basis of such data without compression is ineffective. In addition, when recognizing on the basis of pairwise comparison of objects with each other, it is often difficult to systematize the results. The proposed recognition method is aimed at solving these problems. In recognition, spectral de-scriptors are applied using characteristics that describe various physical processes on the surface. These descriptors use the spectral decomposition of a discrete analogue of the Laplace-Beltrami operator for objects whose surface is approximated by a triangular grid. Each of the presented objects is represented by three descriptors. A method of compressing information about the shape of an object represented by descriptors using the entered spectral distribution maps is proposed. The peculiarity of this compression method is that when it is used it: is possible to compare objects of different levels of detail, the recognition process is accelerated, and the important properties of resistance to noise and invariance to various form transformations possessed by spectral descriptors are preserved. Object recognition is then performed using deep machine learning, which uses a convolutional neural network with several channels. The input data for each channel of the neural network are maps of spectral distributions. Recognition is performed by computing in a pre-trained neural network and then determining the class to which the object belongs. A series of computational experiments using various configurations of spectral descriptors was carried out. Experimental results demonstrate high recognition accuracy for three-dimensional objects with various shape transformations.
References
2. Waleed M., Ben Hamza A. Reeb graph path dissimilarity for 3D object matching and retrieval // The Visual Computer. Springer. — 2012. — Vol. 28. — P. 305-318.
3. Macrini D., Siddiqi K., Dickinson S. From Skeletons to Bone Graphs: Medial Abstraction for Object Recognition // 26th IEEE Conference on Computer Vision and Pattern Recognition, CVPR. Anchorage, Alaska, USA. — 2008. — P. 1-8.
4. Ringler T., Randall D. A Potential Enstrophy and Energy Conserving Numerical Scheme for Solution of the Shallow-Water Equations on a Geodesic Grid // American Meteorological So-ciety. — 2002. — Vol. 130. — P. 1397-1410.
5. Chuang J., Tsai C., Ko M. Skeletonization of three-dimensional object using generalized poten-tial field // IEEE Transactions on Pattern Analysis and Machine Intelligence. — 2000. — Vol. 22(11). — P. 1241-1251.
6. Местецкий Л.М., Зимовнов А.В. Построение криволинейного скелета трехмерной модели по плоским проекциям. // Вестник ТвГУ. Серия: Прикладная математика. — 2016 . — №3. — C. 67-83.
7. Varun J., Zhang H. A spectral approach to shape-based retrieval of articulated 3D models // Computer-Aided Design. — 2007. —Vol. 39(5). — P. 398-407.
8. Chung F.R.K, Spectral graph theory // American Mathematical Society. —1997.
9. Reuter M., Wolter F.-E., Peinecke N. Laplace-spectra as fingerprints for shape matching // Proceedings of the Ninth ACM Symposium on Solid and Physical Modeling, Cambridge, Mas-sachusetts, USA. — 2005. — P. 101-106.
10. Rustamov M. Laplace-Beltrami eigenfunctions for deformation invariant shape representation // Proceedings of symposium on geometry processing, Barcelona, Spain. — 2007. — P. 225-233.
11. Sun J., Ovsjanikov M., Guibas L.J. A concise and provably informative multi-scale signature based on heat diffusion // Comput. Graph. Forum. — 2009 Vol. 28(5). — P. 1383-1392.
12. Aubry M., Schlickewei U., Cremers D.The wave kernel signature: a quantum mechanical ap-proach to shape analysis // IEEE International Conference on Computer Vision Workshops, ICCV. Barcelona, Spain. — 2011 . — P. 1626-1633.
13. Hammond D.K., Vandergheynst P., Gribonval R. Wavelets on graphs via spectral graph theory // Applied and Computational Harmonic Analysis. — 2011. — Vol. 30(2). — P. 129-150.
14. LeCun Y., Bottou L., Bengio Y., Haffner P. Gradient-based learning applied to document recog-nition // Proceedings of the IEEE. — 1999. — Vol. 86(11). — P. 2278-2324.
15. Krizhevsky A., Sutskever I., Hinton G.E. ImageNet classification with deep convolutional neural networks // NIPS'12 Proceedings of the 25th International Conference on Neural Information Processing Systems, Lake Tahoe, USA. — 2012. — P. 1097-1105.
16. Sitnik R., Karaszewski M. Optimized point cloud triangulation for 3D scanning systems // Ma-chine Graphics and Vision . — 2008. — Vol. 17(4). — P. 349-371.
17. Kobbelt L.P., Botsch M. An Interactive Approach to Point Cloud Triangulation // Computer Graphics Forum. — 2000. — Vol. 19(3). — P. 479-487.
18. Курант Р., Гильберт Д. Методы математической физики // Издательство "Мир", Москва, т. 2. —1951. — 1020 С.
19. Chavel I., Riemannian Geometry: A Modern Introduction // Cambridge University Press, Cambridge. — 2006. — Vol. 98. — P. 471.
20. Meyer M., Desbrun M., Schroder P., Barr A.Discrete differential geometry operators for trian-gulated 2-manifolds // In Proceedings of Visual Mathematics. — 2002. — P. 35-57.
21. Xu G. Discrete Laplace-Beltrami operator on sphere and optimal spherical triangulations // Int. J. Comput. Geometry Appl. — 2006. — Vol. 16(1). — P. 75-93.
22. McGill DATABASE. — URL: www.cim.mcgill.ca/~shape/benchMark/.
