Indexation d’Objets 3D par Spectre de Formes Géodésiques 3D (SFG3D)

Indexation d’Objets 3D par Spectre de Formes Géodésiques 3D (SFG3D)

Faten Chaieb Wieme Gadacha  Faouzi Ghorbel 

Laboratoire CRISTAL, ENSI, Université La Manouba Campus Universitaire de La Manouba, 2010, Tunisie

Page: 
279-291
|
DOI: 
https://doi.org/10.3166/TS.31.279-291
Received: 
N/A
| |
Accepted: 
N/A
| | Citation

OPEN ACCESS

Abstract: 

In this paper, we address the problem of 3D object retrieval based on 3D shape descriptors. The proposed approach builds a new descriptor intrinsically invariant to geometric transformations and robust to topology changes and remeshing. The 3D shape spectrum Descriptor (3D SSD), proposed in the MPEG-7 (Zaharia, 2001), is computed on an intrinsic interest point neighborhood. The neighborhood around each interest point is composed of a set of geodesic level curves and radial ones. The level curves correspond to the points at equal geodesic distances from the interest point. The experiments carried out on the SHREC’09 and SHREC’11 datasets show the performance of the proposed descriptor and compare it to further descriptors proposed in the literature (Lian, 2009;2011). 

Extended Abstract 

The continuous development of multimedia technologies and virtual reality has led to an increasing interest in the use of three-dimensional content (3D) in many applications such as medical imaging, games, archeology, etc. The growth of such data requires efficient and rapid content-based 3D shape retrieval systems. To achieve this idea, shape descriptors can be used to provide a unique, compact and significant content description of a 3D object. These descriptors should verify invariance properties with respect to a class of rigid and non-rigid transformations, the robustness to topology changes and stability (insensitivity to noise). 

Several 3D-model retrieval approaches and shape descriptors have been introduced in the literature (Zaharia, 2001; Wang, 2007; Kazhdan, 2003; Novotni, 2003; Vranic, 2001; Lian, 2013). Global descriptors describe the geometrical properties of the object in its totality. However, they can only discriminate between general categories. Local descriptors describe the local surface region around a number of sample points on an object. They are known to be sensitive to non-rigid transformations of 3D objects.  

To deal with the problem non-rigid objects retrieval, several methods based on local descriptors have been proposed (Lian et al. 2013). In this context, we propose a 3D shape description approach which calculates global descriptors in a geodesic neighborhood around interest points (see figure 1).

The first step is to refine the initial 3D mesh by Mid-Edge subdivision scheme (see Figure 2). This mesh subdivision step will provide a smoother mesh and will make the description more robust to various topological representations. Then, a set of 3D interest points is extracted using the 3D Harris detector (Sipiran and Bustos, 2011) which is an extension of 2D images Harris (Harris and Stephens, 1988). Then a geodesic neighborhood is defined around each interest point. It consists of a set of iso-geodesic closed curves and radial lines. An iso-geodesic closed curve corresponds to vertices located at the same geodesic distance from the interest point. Radial curves segment the neighborhood into a set of sectors. The intersection of radial lines and iso-geodesic curves provide a set of patches (connected set of adjacent triangles). Finally, the 3D shape spectrum is calculated for each patch. 

The proposed descriptor, called geodesic spectrum shape or 3D SSD-Harris3D was evaluated and tested on a variety of 3D objects data sets. In fact, the experiments carried out on the SHREC’09 and SHREC’11 datasets show the performance of the proposed descriptor and compare it to further descriptors proposed in the literature (Lian, 2009; 2011). Obtained results are promising and show the interest of this approach. 

Furthermore, the neighborhood size (number of iso-geodesic curves and radial lines), and its resolution (the distance between curves) affect the accuracy and quality of the proposed descriptor. Therefore, further study on curves resolution will be the subject of future work. Also, the geodesic neighborhood will define a local parameterization around each interest point which will motivate the use of other global descriptors such as Fourier descriptors. 

RÉSUMÉ

Dans ce papier, nous nous intéressons au problème d’indexation d’objets 3D par descripteurs de formes. L’approche proposée consiste à construire un nouveau descripteur intrinsèquement invariant aux transformations géométriques et robuste aux changements de topologie et au remaillage. Il s’agit de calculer au voisinage des points d’intérêts de la surface externe de la forme, le spectre de forme 3D (SF3D) retenu dans le standard MPEG-7 (Zaharia, 2001). Le voisinage autour de chaque point d’intérêt est obtenu par intersection des courbes de niveaux géodésiques et des lignes radiales. Les courbes de niveaux correspondent aux points situés à égales distances géodésiques du point d’intérêt. Les expériences effectuées sur les bases de test de SHREC’09 et de SHREC’11 montrent les propriétés du descripteur proposé et valident la pertinence de notre approche par rapport aux descripteurs proposés dans la littérature (Lian, 2009 ;2011). 

Keywords: 

 3D retrieval, 3D shape descriptor, geodesic neighborhood, similarity distance, hausdroff distance, interest point. 

MOTS-CLÉS

indexation 3D, descripteur de formes 3D, voisinage geodesique, mesure de similarité, distance hausdroff, point d’intérêt. 

1. Introduction
2. Approche Proposée
3. Mesure de Similarité
4. Résultats Expérimentaux
5. Conclusion et Perspectives
  References

Bronstein A. M., Bronstein M. M., Kimmel R. (2008). Numerical geometry of non-rigid shapes. Springer-Verlag, New York. 

Cohen L.D., Kimmel R. (1997). Global minimum for active contour model: A minimal path approach. International Journal of Computer Vision, vol. 24, n° 1, p. 57-78. 

Harris C., Stephens M. (1988). A combined corner and edge detection. Proceedings of Fourth Alvey Vision Conference.p. 147-151. 

Kazhdan M., Funkhouser T., Rusinkiewicz S. (2003). Rotation invariant spherical harmonic representation of 3D shape descriptors. Symposium on Geometry Processing. Lavoisier. 

Laganière R. (1998). A morphological operator for corner detection. Pattern Recognition, vol. 31, n° 11, p. 1643-1652. 

Lavoué G. (2011).Bag of words and local spectral descriptor for 3D partial shape retrieval. Proceedings of the Eurographics Workshop on 3D Object Retrieval (3DOR11), p. 41-48. 

Lian Z., Godil A., Bustos B., Daoudi M., Hermans J., Kawamura S., Kurita Y., Lavoué G., Nguyen H.V., Ohbuchi R., Ohkita Y., Reuter M., Sipiran I., Smeets D., Suetens P., Tabia H., Vandermeulen D. (2013). A comparison of methods for non-rigid 3D shape retrieval. Pattern Recognition vol. 46, n° 1. p. 449-461. 

Lian Z., Godil A., Bustos B., Daoudi M., Hermans J., Kawamura S., Kurita Y., Lavoué G., Nguyen H.V., Ohbuchi R., Ohkita Y., Reuter M., Sipiran I., Smeets D., Suetens P., Tabia H., Vandermeulen D. (2011). Shape Retrieval Contest of Non-rigid 3D Watertight Meshes. http ://www.itl.nist.gov/iad/vug/sharp/contest/2011/NonRigid/ 

Lian Z., Godil A., Bustos B., Daoudi M., Hermans J., Kawamura S., Kurita Y., Lavoué G., Nguyen H.V., Ohbuchi R., Ohkita Y., Reuter M., Sipiran I., Smeets D., Suetens P., Tabia H., Vandermeulen D. (2009). Shape Retrieval Contest on a New Generic Shape Benchmark. http ://www.itl.nist.gov/iad/vug/sharp/benchmark/shrecGeneric/ 

Novotni M., Klein R. (2003). 3D Zernike descriptors for content based shape retrieval. ACM SIGGRAPH symposium on Geometry processing, p. 156-164, Switzerland. 

Peyré G., Cohen L. D. (2005). Progress in Nonlinear Differential Equations and Their Applications, Geodesic Computations for Fast and Accurate Surface Remeshing and Parameterization, vol. 63, p. 157-171. Springer.  

Schmid C., Mohr R. (1997). Local gray level value invariant for image retrieval, IEEE Trans. Pattern Anal. Match. Intell. vol. 19, n° 5, p. 530-535.  

Sipiran I., Bustos B. (2011). Harris 3D: a robust extension of the Harris operator for interest point detection on 3D meshes. The Visual Computer, vol. 27, n° 11. 

Smeets D., Fabry T., Hermans J., Vandermeulen D., Suetens P. (2009). Isometric deformation modelling for object recognition. CAIP’09: Int. Conf. Computer Analysis of Images and Patterns, LNCS 5702, p. 757-765. 

Vrani D.V., Saupe D. (2001). 3D shape descriptor based on 3D Fourier transform. EURASIP Conference on Digital Signal Processing for Multimedia Communications and Services (ECMCS’01), p. 271- 274, Budapest. 

Wang D., Zhang J., Wong H., Li Y. (2007).3D model retrieval based on multi-shell extended gaussian image. Advances in Visual Information Systems, p. 426-437, Shanghai, China. 

Zaharia T.B. (2001). Indexation de vidéos et de maillages 3D dans le contexte MPEG-7. Thèse en Informatique, Université René Descartes, Paris V.