## Retrieving articulated 3-D models using medial surfaces (2008)

Citations: | 39 - 3 self |

### BibTeX

@MISC{Siddiqi08retrievingarticulated,

author = {Kaleem Siddiqi and Juan Zhang and Diego Macrini and Ali Shokoufandeh and Sylvain Bouix and Sven Dickinson},

title = { Retrieving articulated 3-D models using medial surfaces},

year = {2008}

}

### OpenURL

### Abstract

We consider the use of medial surfaces to represent symmetries of 3-D objects. This allows for a qualitative abstraction based on a directed acyclic graph of components and also a degree of invariance to a variety of transformations including the articulation of parts. We demonstrate the use of this representation for 3-D object model retrieval. Our formulation uses the geometric information

### Citations

5216 | Distinctive image features from scale-invariant keypoints - Lowe |

2814 | Eigenfaces for Recognition - Turk, Pentland - 1990 |

962 | Visual learning and recognition of 3d objects from appearance - Murase, Nayar - 1995 |

750 | Recognition-by-components: A theory of human image understanding
- Biederman
- 1987
(Show Context)
Citation Context ...er vision community has grappled with the problem of generic level object recognition by suggesting representations based on volumetric parts, including generalized cylinders, superquadrics and geons =-=[5, 16, 19, 4]-=-. Such approaches build a degree of robustness to deformations and movement of parts, but their representational power is limited by the vocabulary of geometric primitives that are selected. Motivated... |

683 |
Principles of Categorization
- Rosch
- 1978
(Show Context)
Citation Context ...y at coarser levels, e.g., the classification human. In the context of generic 3-D model retrieval, such coarser levels in fact correspond to the notion of a basic level or entry level categorization =-=[21, 4]-=-, whose exemplars might reflect a variety of complex poses and articulations, such as those seen in Fig. 1. Our matching and indexing algorithms have the potential to work at this more challenging lev... |

452 |
Representation and recognition of the spatial organization of threedimensional shapes
- Marr, HK
- 1978
(Show Context)
Citation Context ...er vision community has grappled with the problem of generic level object recognition by suggesting representations based on volumetric parts, including generalized cylinders, superquadrics and geons =-=[5, 16, 19, 4]-=-. Such approaches build a degree of robustness to deformations and movement of parts, but their representational power is limited by the vocabulary of geometric primitives that are selected. Motivated... |

378 | The SR-tree: An Index Structure for High-Dimensional Nearest Neighbor Queries - Katayama, Satoh - 1997 |

303 |
T.L.: Topology matching for fully automatic similarity estimation of 3D shapes
- Hilaga, Shinagawa, et al.
(Show Context)
Citation Context ...ility distributions representing 3D object surfaces. In the domain of graph theory there have also been attempts to address the problem of 3D shape matching using representations based on Reeb graphs =-=[24, 10]-=-. These allow for topological properties to be captured, at least in a coarse sense. An alternative approach is to use 3-D medial loci (3-D skeletons), obtained by considering the locus of centers of ... |

223 | R.C.: Three-dimensional object recognition - Besl, Jain - 1985 |

208 | The princeton shape benchmark
- Shilane, Min, et al.
- 2004
(Show Context)
Citation Context ...al size normalization. 5 Experimental Results In order to test the power of our indexing and matching algorithms using medial surfacebased DAGs, we have considered using the Princeton Shape Benchmark =-=[23]-=-. This standardized database, which contains 1,814 3-D object models organized by class, is an effective one for comparing the performance of a variety of methods including those in [11, 17, 3, 30, 7]... |

206 | Rotation invariant spherical harmonic representation of 3D shape descriptors
- Kazhdan, Funkhouser, et al.
- 2003
(Show Context)
Citation Context ...e along with an eigenvalue labeling of the adjacency matrix of the subgraph rooted at that node. We present comparative results against the techniques of shape distributions [17] and harmonic spheres =-=[12]-=- on a database of 320 models representing 13 object classes. The results demonstrate that medial surface based graph matching significantly outperforms these techniques for objects with articulating p... |

203 | The power crust
- Amenta, Choi, et al.
- 2001
(Show Context)
Citation Context ... surfaces and their graph spectra in the context of 3-D object model retrieval and indexing. 2 Medial Surfaces and DAGs Recent approaches for computing 3-D skeletons include the power crust algorithm =-=[2]-=-, the shock scaffold [13] and average outward flux-based skeletons [27]. The first two methods have the advantage that they can be employed on input data in the form of points sampled from an object’s... |

199 | Distance transformations in arbitrary dimensions - Borgefors - 1984 |

194 | Shape distributions
- Osada, Funkhouser, et al.
- 2002
(Show Context)
Citation Context ...n associated with each node along with an eigenvalue labeling of the adjacency matrix of the subgraph rooted at that node. We present comparative results against the techniques of shape distributions =-=[17]-=- and harmonic spheres [12] on a database of 320 models representing 13 object classes. The results demonstrate that medial surface based graph matching significantly outperforms these techniques for o... |

183 | Modal matching for correspondence and recognition - Sclaroff, Pentland - 1995 |

176 |
Visual perception by computer
- Binford
- 1971
(Show Context)
Citation Context ...er vision community has grappled with the problem of generic level object recognition by suggesting representations based on volumetric parts, including generalized cylinders, superquadrics and geons =-=[5, 16, 19, 4]-=-. Such approaches build a degree of robustness to deformations and movement of parts, but their representational power is limited by the vocabulary of geometric primitives that are selected. Motivated... |

172 | The power crust, unions of balls, and the medial axis transform - Amenta, Choi, et al. - 2001 |

171 | Matching hierarchical structures using association graphs - Pelillo, Siddiqi, et al. - 1999 |

161 | Model-based recognition in robot vision - Chin, Dyer - 1986 |

159 |
Perceptual organization and the representation of natural form
- Pentland
- 1986
(Show Context)
Citation Context |

151 | Recognition of shapes by editing their shock graphs
- Sebastian, Klein, et al.
- 2004
(Show Context)
Citation Context ...atching 3-D object models. These and related algorithms have already shown promise in the computer vision community for generic level view-based object indexing and matching using 2-D skeletal graphs =-=[28, 25, 18, 22]-=-. They have also been demonstrated in the context of matching 3-D object models with tubular parts, using a centerline approximation of the 3-D skeleton [29]. We demonstrate their significant potentia... |

151 | A survey of free-form object representation and recognition techniques - Campbell, Flynn |

145 |
Biological Shape and Visual Science
- Blum
- 1973
(Show Context)
Citation Context ...ptured, at least in a coarse sense. An alternative approach is to use 3-D medial loci (3-D skeletons), obtained by considering the locus of centers of maximal inscribed spheres along with their radii =-=[6]-=-. As pointed out by Blum, this offers the advantage that a graph of parts can be inferred from the underlying local mirror symmetries of the object. To motivate this idea, consider once again the huma... |

141 | T.: 3D shape histograms for similarity search and classification in spatial databases
- Ankerst, Kastenmüller, et al.
- 1999
(Show Context)
Citation Context ...hanges of pose. For such examples, the very notion of a center of mass or a rigid reference point[1], which is crucial for the computation of descriptions such as shape histograms (sectors or shells) =-=[3]-=- or spherical extent functions [30], can be nonintuitivesFig. 1. Exemplars of the object class “human” created by changes in pose and articulations of parts (top row). The medial surface (or 3-D skele... |

123 | The hamilton-jacobi skeleton
- Siddiqi, Bouix, et al.
- 1999
(Show Context)
Citation Context ...ntuitivesFig. 1. Exemplars of the object class “human” created by changes in pose and articulations of parts (top row). The medial surface (or 3-D skeleton) of each is computed using the algorithm of =-=[27]-=- (bottom row). The medial surface is automatically partitioned into distinct parts, each shown in a different color. and arbitrary. In fact, the centroid of such models may actually lie in the backgro... |

100 | Skeleton based shape matching and retrieval
- Sundar, Silver, et al.
- 2003
(Show Context)
Citation Context ...hing using 2-D skeletal graphs [28, 25, 18, 22]. They have also been demonstrated in the context of matching 3-D object models with tubular parts, using a centerline approximation of the 3-D skeleton =-=[29]-=-. We demonstrate their significant potential for medial surface-based 3-D object retrieval with experimental results on a database of 320 models representing 13 object classes, including exemplars of ... |

95 | Content Based Retrieval of VRML objects - an iterative and interactive approach
- Elad, Tal, et al.
- 2001
(Show Context)
Citation Context ...lie in the background. To complicate matters, it is unclear how to obtain a global alignment of such models. As well, measures based on reflective symmetries [11], and signatures based on 3-D moments =-=[7]-=- or chord histograms [17] are not invariant under such transformations. The computer vision community has grappled with the problem of generic level object recognition by suggesting representations ba... |

82 |
Surface coding based on morse theory
- Shinagawa, Kunii, et al.
- 1991
(Show Context)
Citation Context ...ility distributions representing 3D object surfaces. In the domain of graph theory there have also been attempts to address the problem of 3D shape matching using representations based on Reeb graphs =-=[24, 10]-=-. These allow for topological properties to be captured, at least in a coarse sense. An alternative approach is to use 3-D medial loci (3-D skeletons), obtained by considering the locus of centers of ... |

76 | D.: 3d model retrieval with spherical harmonics and moments
- Vranic, Saupe
(Show Context)
Citation Context ... the very notion of a center of mass or a rigid reference point[1], which is crucial for the computation of descriptions such as shape histograms (sectors or shells) [3] or spherical extent functions =-=[30]-=-, can be nonintuitivesFig. 1. Exemplars of the object class “human” created by changes in pose and articulations of parts (top row). The medial surface (or 3-D skeleton) of each is computed using the ... |

60 | S.: A reflective symmetry descriptor for 3D models
- Kazhdan, Chazelle, et al.
- 2003
(Show Context)
Citation Context ... the centroid of such models may actually lie in the background. To complicate matters, it is unclear how to obtain a global alignment of such models. As well, measures based on reflective symmetries =-=[11]-=-, and signatures based on 3-D moments [7] or chord histograms [17] are not invariant under such transformations. The computer vision community has grappled with the problem of generic level object rec... |

56 | Distance-Ordered Homotopic Thinning: A Skeletonization Algorithm for 3D - Pudney - 1998 |

53 |
A formal classification of 3D medial axis points and their local geometry
- Giblin, Kimia
(Show Context)
Citation Context ...d to characterize 3-D skeletal points. We choose to employ this latter method since it has the advantage that the digital classification of [15] allows for the taxonomy of generic 3-D skeletal points =-=[8]-=- to be interpreted on a rectangular lattice, leading to a graph of parts. Under the assumption that the initial model is given in triangulated form, we begin by scaling all the vertices so that they f... |

51 | Generic model abstraction from examples - Keselman, Dickinson |

50 | Topological segmentation of discrete surfaces
- Malandain, Bertrand, et al.
- 1993
(Show Context)
Citation Context ...Euclidean distance function gradient vector field can be used to characterize 3-D skeletal points. We choose to employ this latter method since it has the advantage that the digital classification of =-=[15]-=- allows for the taxonomy of generic 3-D skeletal points [8] to be interpreted on a rectangular lattice, leading to a graph of parts. Under the assumption that the initial model is given in triangulate... |

49 | Indexing using a spectral encoding of topological structure
- Shokoufandeh, Dickinson, et al.
- 1999
(Show Context)
Citation Context ...atching 3-D object models. These and related algorithms have already shown promise in the computer vision community for generic level view-based object indexing and matching using 2-D skeletal graphs =-=[28, 25, 18, 22]-=-. They have also been demonstrated in the context of matching 3-D object models with tubular parts, using a centerline approximation of the 3-D skeleton [29]. We demonstrate their significant potentia... |

49 | Automatic Generation of Object Recognition Programs - Ikeuchi, Kanade - 1988 |

43 | Indexing hierarchical structures using graph spectra
- Shokoufandeh, Macrini, et al.
- 2005
(Show Context)
Citation Context ... recent technique to compute medial surfaces [27] by proposing an interpretation of its output as a directed acyclic graph (DAG) of parts. We then use refinements of algorithms based on graph spectra =-=[26]-=- to tackle the problems of indexing and matching 3-D object models. These and related algorithms have already shown promise in the computer vision community for generic level view-based object indexi... |

40 | Matching shapes with a reference point
- Alt, Aichholzer, et al.
- 1994
(Show Context)
Citation Context ... models in Fig. 1. These four exemplars of an object class were created by articulations of parts and changes of pose. For such examples, the very notion of a center of mass or a rigid reference point=-=[1]-=-, which is crucial for the computation of descriptions such as shape histograms (sectors or shells) [3] or spherical extent functions [30], can be nonintuitivesFig. 1. Exemplars of the object class “h... |

40 |
An analysis of a good algorithm for the subtree problem
- Reyner
- 1977
(Show Context)
Citation Context ...was small. In fact, such a formulation amounts to finding the maximum cardinality, minimum weight matching in a bipartite graph spanning the two sets of nodes. In a modification of Reyner’s algorithm =-=[20]-=-, we combine the above bipartite matching formulation with a greedy, best-first search in a recursive procedure to compute the corresponding nodes in two rooted DAGs which, in turn, yields an overall ... |

34 | Medial representations: Mathematics, algorithms and applications - Siddiqi, Pizer - 2008 |

29 | 3D object retrieval using many-to-many matching of curve skeletons - Cornea, Demirci, et al. - 2005 |

29 | Origanizing Large Structural Modelbases - Segupta, Boyer - 1995 |

28 | Flux invariants for shape - Dimitrov, Damon, et al. |

27 | Subgraph isomorphism in polynomial time - MESSMER, BUNKE - 1995 |

23 | Geodesic Object Representation and Recognition
- Hamza, Krim
- 2003
(Show Context)
Citation Context ...information using probabilistic descriptors. These allow intrinsic geometric information to be captured by low dimensional signatures. An elegant example of this is the geodesic shape distribution of =-=[9]-=- where information theoretic measures are used to compare probability distributions representing 3D object surfaces. In the domain of graph theory there have also been attempts to address the problem ... |

21 | On the representation and matching of qualitative shape at multiple scales - Shokoufandeh, Dickinson, et al. |

20 | Biological shape and visual - Blum - 1973 |

19 | Learning hierarchical shape models from examples - Levinshtein, Sminchisescu, et al. - 2005 |

19 | ékely. Characterization and recognition of 3d organ shape in medical image analysis using skeletonization - Näf, Kübler, et al. - 1996 |

19 | Model indexing: the graph-hashing approach - Sossa, Horaud - 1992 |