## From Regular Images to Animated Heads: A Least Squares Approach (1998)

Venue: | in European Conference on Computer Vision |

Citations: | 23 - 8 self |

### BibTeX

@INPROCEEDINGS{Fua98fromregular,

author = {P. Fua and C. Miccio},

title = {From Regular Images to Animated Heads: A Least Squares Approach},

booktitle = {in European Conference on Computer Vision},

year = {1998},

pages = {188--202},

publisher = {Springer}

}

### OpenURL

### Abstract

We show that we can effectively fit arbitrarily complex animation models to noisy image data. Our approach is based on least-squares adjustment using of a set of progressively finer control triangulations and takes advantage of three complementary sources of information: stereo data, silhouette edges and 2-D feature points.

### Citations

3277 | Snakes: active contour models
- Kass, Witkin, et al.
- 1988
(Show Context)
Citation Context ...t optimizer consistently yields better minima and needs fewer iterations at each level than other methods such as the Euler-Lagrange methods that have been extensively used in the Computer Vision eld =-=[13]-=-|the so-called snake approaches|and one of the best known implementations of a quasi-newton approach in the eld of Operations Research [16]. 3 From Image Data to Observations In this section we focus ... |

1544 |
Practical Optimization
- Gill, Murray, et al.
- 1981
(Show Context)
Citation Context ...ies of the algorithm. This is in contrast to typical optimization approaches whose convergence properties tend to degrade when using an objective function that is the sum of many incommensurate terms =-=[10, 9]-=-. These attributes of our approach allow us to derive convincing models even from relatively low-resolution images. We typically start with a set of stereo image pairs or a video sequence. In this wor... |

373 | Camera self-calibration: theory and experiments
- Faugeras, Luong, et al.
- 1992
(Show Context)
Citation Context ... because there are well established photogrammetric techniques, such as bundle-adjustment, that allow the computation of these models as needed. Furthermore recent work in the area of autocalibration =-=[6, 21, 17]-=- could be brought to bear to automate the process. We then go through the following three steps: { We compute disparity maps for each stereo pair or each consecutive pair in the video sequences, t loc... |

326 | Realistic Modeling for Facial Animation
- Lee, Terzopoulos, et al.
- 1995
(Show Context)
Citation Context ...odels with large numbers of degrees of freedom that are suitable for animation purposes but require very clean data{ the kind produced by a laser scanner or structured light|to instantiate them (e.g. =-=[15]-=-). 1 Turing Institute, GlasgowsThe approach proposed here bridges the gap between these two classes by tting to actual image data a detailed face model that has successfully been used to animate virtu... |

225 | Autocalibration and the absolute quadric, in
- Triggs
- 1997
(Show Context)
Citation Context ... because there are well established photogrammetric techniques, such as bundle-adjustment, that allow the computation of these models as needed. Furthermore recent work in the area of autocalibration =-=[6, 21, 17]-=- could be brought to bear to automate the process. We then go through the following three steps: { We compute disparity maps for each stereo pair or each consecutive pair in the video sequences, t loc... |

114 | The Integration of Optical Flow and Deformable Models with Applications to Human Face Shape and Motion Estimation
- DeCarlo, Metaxas
- 1996
(Show Context)
Citation Context ...nto the following two categories: { Some concentrate on tracking the head motion and some features. They typically use a fairly coarse face model that is too simple for realistic face animation (e.g. =-=[4]-=-). { Others use sophisticated face models with large numbers of degrees of freedom that are suitable for animation purposes but require very clean data{ the kind produced by a laser scanner or structu... |

87 |
MINOS 5.4 User’s Guide
- Murtagh, Saunders
- 1995
(Show Context)
Citation Context ...s that have been extensively used in the Computer Vision eld [13]|the so-called snake approaches|and one of the best known implementations of a quasi-newton approach in the eld of Operations Research =-=[16]-=-. 3 From Image Data to Observations In this section we focus on stereo range data and silhouettes because they can be readily acquired from image sequences and form two complementary sources of inform... |

79 | Parametrized Structure from Motion for 3D Adaptive Feedback Tracking of Faces
- Jebara, Pentland
- 1997
(Show Context)
Citation Context ...ntee that important features of the model|mouth and corners of the eyes especially|project at the correct locations in the face. Because there now are automated techniques to e ectively extract those =-=[2, 3, 11]-=-, a system likes ours must be able to take advantage of those whenever available. For each vertex xi;yi;zi of the surface triangulation whose 2{D projection ui;vi is known, we can write two observatio... |

68 | 3D position, attitude and shape input using video tracking of hands and lips”, Computer Graphics (Proc
- Blake, Isard
- 1994
(Show Context)
Citation Context ...ntee that important features of the model|mouth and corners of the eyes especially|project at the correct locations in the face. Because there now are automated techniques to e ectively extract those =-=[2, 3, 11]-=-, a system likes ours must be able to take advantage of those whenever available. For each vertex xi;yi;zi of the surface triangulation whose 2{D projection ui;vi is known, we can write two observatio... |

66 |
From multiple stereo views to multiple 3D surfaces
- Fua
- 1997
(Show Context)
Citation Context ... surface information for arbitrary surface orientation and topology. Figure 2(e) depicts the result of this procedure. For additional details, we refer the interested reader to an earlier publication =-=[8]-=-. The center of each patch is treated as an attractor. The easiest way to handle is to model it as a spring attached to the mesh vertex closest to it. This, however, is inadequate if one wishes to use... |

65 | Simulation of facial muscle actions based on rational free form deformation
- Kalra, Mangili, et al.
- 1992
(Show Context)
Citation Context ... reconstruction and animation results on a number of di erent heads.s2 Least Squares Framework In this work, we use the facial animation model that has been developed at University of Geneva and EPFL =-=[12]-=-. It can produce the di erent facial expressions arising from speech and emotions. Its multilevel con guration reduces complexity and provides independent control for each level. At the lowest level, ... |

52 | The direct computation of height from shading
- Leclerc, Bobick
- 1991
(Show Context)
Citation Context ...nders. In recent years much work has been devoted to the modelling of faces from image and range data. There are many e ective approaches to recovering face geometry. They rely on stereo [5], shading =-=[14]-=-, structured light [19], silhouettes [20] orlow-intensity lasers. Some of these systems such as the C3D tm1 or the Cyberware tm scanner are commercially available. However, recovering a head as a simp... |

33 |
Numerical Recipes: The Art of Scienti® c Computing (Cambridge
- Press, Flannery, et al.
- 1986
(Show Context)
Citation Context ...deviation from the model. We will then minimize X 1 i nobs wi 2 i ; (2) where wi is a weight associated to each observation. The optimization is then performed using the Levenberg-Marquardt algorithm =-=[18]-=-. In theory we could take the parameter vector P to be the vector of all x; y, and z coordinates of the surface triangulation. However, because the image data is very noisy, wewould have to impose a v... |

23 |
A Parallel Stereo Algorithm that
- Fua
- 1993
(Show Context)
Citation Context ...onochrome and registered so that their relative camera models are known a priori. Since we are interested in reconstructing surfaces, we start the process by using a simple correlationbased algorithm =-=[7]-=- to compute a disparity map for each pair or triplet and by turning each valid disparity value into a 3{D point. If other sources of range data were available, they could be used in a similar fashion.... |

18 |
Imposing Hard Constraints on Deformable Models through Optimization
- Fua, Brechbuehler
- 1997
(Show Context)
Citation Context ...ies of the algorithm. This is in contrast to typical optimization approaches whose convergence properties tend to degrade when using an objective function that is the sum of many incommensurate terms =-=[10, 9]-=-. These attributes of our approach allow us to derive convincing models even from relatively low-resolution images. We typically start with a set of stereo image pairs or a video sequence. In this wor... |

17 |
Active acquisition of 3d shape for moving objects
- Proesmans, VanGool
- 1996
(Show Context)
Citation Context ...much work has been devoted to the modelling of faces from image and range data. There are many e ective approaches to recovering face geometry. They rely on stereo [5], shading [14], structured light =-=[19]-=-, silhouettes [20] orlow-intensity lasers. Some of these systems such as the C3D tm1 or the Cyberware tm scanner are commercially available. However, recovering a head as a simple triangulated mesh do... |

15 |
Locating Objects of Varying Shape Using Statistical Feature Detectors
- Cootes, Taylor
- 1996
(Show Context)
Citation Context ...ntee that important features of the model|mouth and corners of the eyes especially|project at the correct locations in the face. Because there now are automated techniques to e ectively extract those =-=[2, 3, 11]-=-, a system likes ours must be able to take advantage of those whenever available. For each vertex xi;yi;zi of the surface triangulation whose 2{D projection ui;vi is known, we can write two observatio... |

14 |
Computing di erential properties of 3-D shapes from stereoscopic images without 3-D models
- Devernay, Faugeras
- 1994
(Show Context)
Citation Context ... laser range nders. In recent years much work has been devoted to the modelling of faces from image and range data. There are many e ective approaches to recovering face geometry. They rely on stereo =-=[5]-=-, shading [14], structured light [19], silhouettes [20] orlow-intensity lasers. Some of these systems such as the C3D tm1 or the Cyberware tm scanner are commercially available. However, recovering a ... |

8 |
Self-Calibration and Metric Reconstruction
- Pollefeys, Koch, et al.
- 1998
(Show Context)
Citation Context ... because there are well established photogrammetric techniques, such as bundle-adjustment, that allow the computation of these models as needed. Furthermore recent work in the area of autocalibration =-=[6, 21, 17]-=- could be brought to bear to automate the process. We then go through the following three steps: { We compute disparity maps for each stereo pair or each consecutive pair in the video sequences, t loc... |

8 |
Analysis-based facial expression synthesis
- Tang, Huang
- 1996
(Show Context)
Citation Context ... devoted to the modelling of faces from image and range data. There are many e ective approaches to recovering face geometry. They rely on stereo [5], shading [14], structured light [19], silhouettes =-=[20]-=- orlow-intensity lasers. Some of these systems such as the C3D tm1 or the Cyberware tm scanner are commercially available. However, recovering a head as a simple triangulated mesh does not su ce: To a... |