## 3D face recognition in the presence of expression: A guidance-based constraint deformation approach (2007)

Venue: | In IEEE CVPR |

Citations: | 7 - 3 self |

### BibTeX

@INPROCEEDINGS{Wang073dface,

author = {Yueming Wang and Gang Pan and Zhaohui Wu},

title = {3D face recognition in the presence of expression: A guidance-based constraint deformation approach},

booktitle = {In IEEE CVPR},

year = {2007},

pages = {1}

}

### OpenURL

### Abstract

Three-dimensional human face recognition in the presence of expression is a big challenge, since the shape distortion caused by facial expression greatly weakens the rigid matching. This paper proposes a guidance-based constraint deformation(GCD) model to cope with the shape distortion by expression. The basic idea is that, the face model with non-neutral expression is deformed toward its neutral one under certain constraint so that the distortion is reduced while inter-class discriminative information is preserved. The GCD model exploits the neutral 3D face shape to guide the deformation, meanwhile applies a rigid constraint on it. Both steps are smoothly unified in the Poisson equation framework. The GCD approach only needs one neutral model for each person in the gallery. The experimental results, carried out on the large 3D face databases– FRGC v2.0, demonstrate that our method significantly outperforms ICP method for both identification and authentication mode. It shows the GCD model is promising for coping with the shape distortion in 3D face recognition. 1.

### Citations

1114 | Progressive meshes
- Hoppe
- 1996
(Show Context)
Citation Context ...rvature analysis of surface. For details, see[20]. (2) After that, the mesh simplification is applied on the face mesh to reduce the number of vertices to meet the requirement of efficient computation=-=[8]-=-. About 8000 vertices and 160000 triangles are preserved in every simplified facial mesh in our experiments. (3) Then, the model is placed to a canonical coordinate system by feature points detection.... |

348 | Discrete Differential-Geometry Operators for Triangulated 2-Manifolds
- MEYER, DESBRUN, et al.
- 2003
(Show Context)
Citation Context ...he set of vertices and K encodes the connectivity of the simplicial complex including the vertices, edges and triangles. For A 1-ring mesh is shown in Fig.2. Some notions are introduced here at first =-=[11, 19, 15, 16]-=-. Discrete Scalar Fields and Discrete Vector Fields: A scalar field u on the triangular mesh M is expressed as: u(v) = ∑ φi(v)ui (2) i where φi is the piecewise-linear basis function valued 1 at verte... |

252 | Overview of the face recognition grand challenge
- Phillips, Flynn, et al.
(Show Context)
Citation Context ...rmation between different subjects holds the difference. 4. Experiments 4.1. Data Figure 6. The model examples from FRGC database. Our experiments use the 3D facial data from FRGC database version 2.0=-=[14]-=-, totally 557 subjects and 4950 3D models. Each face model has the expression label and capturing timestamp. All the non-neutral models in FRGC database are adopted as our probe dataset, totally 1538 ... |

137 | Mesh editing with Poisson-based gradient field manipulation
- YU, ZHOU, et al.
(Show Context)
Citation Context ...uidance while important geometric details are locally deformed in a way keeping their appearance characteristics. In this paper, we incorporate the Poisson-based gradient field manipulation technique =-=[21]-=- into the target-guided deformation of facial mesh. 3.2.1 Poission-based Mesh Deformation For a facial triangular mesh M, each coordinate-component of coordinate 3-tuple of its vertices could be treat... |

103 | Three-dimensional face recognition
- Bronstein, Bronstein, et al.
- 2004
(Show Context)
Citation Context ... model is promising for coping with the shape distortion in 3D face recognition. 1. Introduction The three-dimensional human face recognition has received greatly increasing attentions in recent years=-=[17, 3, 6, 9, 10, 22, 1]-=-. Compared with the facial images, which essentially is the projection of the 3D human face onto a 2D planar domain, 3D facial data have more clues for recognition. Utilizing the three-dimensional inf... |

84 |
3D human face recognition using point signature
- Chua, Han, et al.
- 2000
(Show Context)
Citation Context ... model is promising for coping with the shape distortion in 3D face recognition. 1. Introduction The three-dimensional human face recognition has received greatly increasing attentions in recent years=-=[17, 3, 6, 9, 10, 22, 1]-=-. Compared with the facial images, which essentially is the projection of the 3D human face onto a 2D planar domain, 3D facial data have more clues for recognition. Utilizing the three-dimensional inf... |

80 | Expression-invariant 3D face recognition
- Bronstein, Kimmel
- 2003
(Show Context)
Citation Context ...ted the rigid parts of facial surface for matching to cope with expression variation. They mainly use the nose-near region for recognition. The surface matching was achieved by ICP. A.Bronstein et al =-=[2, 3]-=- assumed that the facial surfaces in the presence of expression is an isometric surface which are not stretched by expressions. All 3D facial models were transformed into a canonical form for recognit... |

74 | A survey of approaches and challenges in 3D and multi-modal 3D+2D face recognition
- Bowyer, Chang, et al.
(Show Context)
Citation Context ... model is promising for coping with the shape distortion in 3D face recognition. 1. Introduction The three-dimensional human face recognition has received greatly increasing attentions in recent years=-=[17, 3, 6, 9, 10, 22, 1]-=-. Compared with the facial images, which essentially is the projection of the 3D human face onto a 2D planar domain, 3D facial data have more clues for recognition. Utilizing the three-dimensional inf... |

69 | Discrete multiscale vector field decomposition
- Tong, Lombeyda, et al.
(Show Context)
Citation Context ...he set of vertices and K encodes the connectivity of the simplicial complex including the vertices, edges and triangles. For A 1-ring mesh is shown in Fig.2. Some notions are introduced here at first =-=[11, 19, 15, 16]-=-. Discrete Scalar Fields and Discrete Vector Fields: A scalar field u on the triangular mesh M is expressed as: u(v) = ∑ φi(v)ui (2) i where φi is the piecewise-linear basis function valued 1 at verte... |

52 | Identifying vector fields singularities using a discrete hodge decomposition
- Polthier, Preuss
- 2003
(Show Context)
Citation Context ...he set of vertices and K encodes the connectivity of the simplicial complex including the vertices, edges and triangles. For A 1-ring mesh is shown in Fig.2. Some notions are introduced here at first =-=[11, 19, 15, 16]-=-. Discrete Scalar Fields and Discrete Vector Fields: A scalar field u on the triangular mesh M is expressed as: u(v) = ∑ φi(v)ui (2) i where φi is the piecewise-linear basis function valued 1 at verte... |

36 | Deformation Modeling for Robust 3D Face Matching
- Lu, Jain
- 2006
(Show Context)
Citation Context |

34 |
Adaptive rigid multi-region selection for handling expression variation in 3d face recognition
- Chang, Bowyer, et al.
(Show Context)
Citation Context ...sed Gaussian distribution to extract rigid parts of facial surface for matching. They recognize 3D face using a local descriptor, named ”Point Signature”, combined with ”voting method”. K.Chang et al =-=[4]-=- also extracted the rigid parts of facial surface for matching to cope with expression variation. They mainly use the nose-near region for recognition. The surface matching was achieved by ICP. A.Bron... |

24 |
Automatic 3D face verification from range data
- Pan, Wu, et al.
- 2003
(Show Context)
Citation Context ...posture and expression etc. For two rigid facial surfaces, the classical matching procedure is to align the surfaces and define a similarity measure for the aligned shapes, such as Hausdorff distance =-=[13, 18]-=-, RMS(the root of mean squared distance)[5]. However, these approaches are usually not applicable to the non∗ Corresponding author: Gang Pan {gpan@zju.edu.cn} rigid cases, which are mostly caused by e... |

23 | Variational approach to vector field decomposition, Scientific Visualization
- Polthier, Preuß
- 2000
(Show Context)
Citation Context |

17 | Y.: 3D face recognition using mapped depth images
- Pan, Han, et al.
- 2005
(Show Context)
Citation Context ...triangles are preserved in every simplified facial mesh in our experiments. (3) Then, the model is placed to a canonical coordinate system by feature points detection. Using the method proposed by Pan=-=[12]-=-, the nose tip pnt, nose base pnb and direction of symmetrical plane ds of face are calculated to fix the six degrees of freedom of facial surface. (4) Finally, a geodesic-based cropping is performed ... |

16 | Face recognition using optimal linear components of range images
- Srivastava, Liu, et al.
- 2006
(Show Context)
Citation Context |

14 |
Intrinsic parametrizations of surface meshes
- Desbrun, Meyer, et al.
- 2002
(Show Context)
Citation Context ...the other three with various expressions in different degree (randomly selected). To set up the correspondence between two meshes, we parameterize each model to a disk to form a parametrization domain=-=[7]-=-. The statistical distribution of the distortion is based on parametrization domain. As shown in Fig. 5, the parametrization domain is discretized into many cells by evenly segmenting the radius and r... |

14 |
A 2D range Hausdorff approach for 3D face recognition
- Russ, Koch, et al.
- 2005
(Show Context)
Citation Context ...posture and expression etc. For two rigid facial surfaces, the classical matching procedure is to align the surfaces and define a similarity measure for the aligned shapes, such as Hausdorff distance =-=[13, 18]-=-, RMS(the root of mean squared distance)[5]. However, these approaches are usually not applicable to the non∗ Corresponding author: Gang Pan {gpan@zju.edu.cn} rigid cases, which are mostly caused by e... |

11 |
Effects on facial expression in 3D face recognition
- Chang, Bowyer, et al.
(Show Context)
Citation Context ... surfaces, the classical matching procedure is to align the surfaces and define a similarity measure for the aligned shapes, such as Hausdorff distance [13, 18], RMS(the root of mean squared distance)=-=[5]-=-. However, these approaches are usually not applicable to the non∗ Corresponding author: Gang Pan {gpan@zju.edu.cn} rigid cases, which are mostly caused by expression variation. A big challenge in 3D ... |

4 |
3D+2D face recognition by fusion at both feature and decision levels
- Li, Zhao, et al.
- 2005
(Show Context)
Citation Context |

1 |
Automatically cropping facial region from raw point cloud data
- Wang, Pan, et al.
- 2006
(Show Context)
Citation Context ...e only the points information, the surface is reconstructed by triangulation. Then, ROI (region of interest) of the face model is roughly cropped by the curvature analysis of surface. For details, see=-=[20]-=-. (2) After that, the mesh simplification is applied on the face mesh to reduce the number of vertices to meet the requirement of efficient computation[8]. About 8000 vertices and 160000 triangles are... |

1 |
Face recognition using optimal linear components of range images
- Zhang, Razdan, et al.
(Show Context)
Citation Context |