#### DMCA

## Stochastic rigidity: Image registration for nowhere-static scenes. (2001)

### Cached

### Download Links

- [imogen.robots.ox.ac.uk:20000]
- [www.robots.ox.ac.uk]
- DBLP

### Other Repositories/Bibliography

Citations: | 52 - 0 self |

### Citations

4680 |
Multiple View Geometry in Computer Vision
- Hartley
- 2000
(Show Context)
Citation Context ...deo sequences. The first step is to define models that can represent each type of motion. For the camera motion this is straightforward: the large body of literature on the geometry of multiple views =-=[6, 10]-=- provides the convenient abstractions and models of projective geometry. In this paper, we generally employ a 2D perspective mapping (homography) between the image planes. For the stochastic component... |

1537 | Nonlinear component analysis as a kernel eigenvalue problem
- Schölkopf, Smola, et al.
- 1998
(Show Context)
Citation Context ...figure 3, the analysis tends to segment independent motions in the scene, choosing one or two components for each region that is moving. Of course, a more robust decomposition than PCA, perhaps K-PCA =-=[22] o-=-r ICA [19], would enhance this effect and potentially improve performance. 2.2. AR model fitting Presenting the trimmed matrix of coefficients, X, to Neumaier and Schneider’sARfit algorithm [20], we... |

1494 |
Three-Dimensional Computer Vision: a Geometric Viewpoint
- Faugeras
- 1993
(Show Context)
Citation Context ...deo sequences. The first step is to define models that can represent each type of motion. For the camera motion this is straightforward: the large body of literature on the geometry of multiple views =-=[6, 10]-=- provides the convenient abstractions and models of projective geometry. In this paper, we generally employ a 2D perspective mapping (homography) between the image planes. For the stochastic component... |

1204 |
random variables and stochastic processes
- Papoulis, Pillai
- 2002
(Show Context)
Citation Context ...ames is essentially a random process. However, by treating the sequence of images as a set of samples from a multidimensional stochastic time-series, we can learn a stochastic model (e.g. an AR model =-=[16, 23]-=-) of the random process which generated the sequence of images. With a static camera, this stochastic model can be used to extend the sequence arbitrarily in time: driving the model with random noise ... |

997 | III. Alignment by maximization of mutual information
- Viola, Wells
- 1995
(Show Context)
Citation Context ...ion here is similar in spirit to the work on space carving [4, 15] where voxel consistency is measured as deviation from a lighting model. It also has the flavour of the early mutual information work =-=[28]-=- where the error metric for image registration is based on deviation from causality of the scattergram of transformed intensities. 1.2. Time series analysis Computer vision has recently seen increased... |

561 | A theory of shape by space carving
- Kutulakos, Seitz
(Show Context)
Citation Context ...in successive layers are related by a compact time-series model. � so that the difference from a time series model. Replacing ����� the function here is similar in spirit to the work on sp=-=ace carving [4, 15]-=- where voxel consistency is measured as deviation from a lighting model. It also has the flavour of the early mutual information work [28] where the error metric for image registration is based on dev... |

296 | Shadow puppetry
- Brand
- 1999
(Show Context)
Citation Context ...tion is based on deviation from causality of the scattergram of transformed intensities. 1.2. Time series analysis Computer vision has recently seen increased use of the tools of time series analysis =-=[2, 3, 13, 17, 18, 24], which model the temporal e-=-volution of physical systems. A time series is a sequence of vector-valued observations ��� � � ��� � � . One task of time series analysis is to forecast the value of � ���... |

272 | ESSA I.: Video textures
- SCHÖDL, SZELISKI, et al.
(Show Context)
Citation Context ...bitrarily in time: driving the model with random noise results in an infinitely varying sequence of images which always looks like the short input sequence. In this way, we can create “videotextures=-=” [21, 24] w-=-hich can play forever without repetition. With a moving camera, the image generation process comprises two components—a stochastic component generated by the videotexture, and a parametric component... |

245 |
Time Series Analysis and its Application
- Shumway, Stoffer
- 2000
(Show Context)
Citation Context ...ames is essentially a random process. However, by treating the sequence of images as a set of samples from a multidimensional stochastic time-series, we can learn a stochastic model (e.g. an AR model =-=[16, 23]-=-) of the random process which generated the sequence of images. With a static camera, this stochastic model can be used to extend the sequence arbitrarily in time: driving the model with random noise ... |

242 | Computing occluding and transparent motions
- Irani, Rousso, et al.
- 1994
(Show Context)
Citation Context ...age of a crowd applauding at a sports event. In these cases, it is not possible to impose the constraint that world points have similar colour in successive views, so existing registration techniques =-=[1, 5, 9, 11] c-=-annot be applied. Indeed the relationship between a point’s colours in successive frames is essentially a random process. However, by treating the sequence of images as a set of samples from a multi... |

190 | A framework for the robust estimation of optical flow
- Black, Anandan
- 1993
(Show Context)
Citation Context ...age of a crowd applauding at a sports event. In these cases, it is not possible to impose the constraint that world points have similar colour in successive views, so existing registration techniques =-=[1, 5, 9, 11] c-=-annot be applied. Indeed the relationship between a point’s colours in successive frames is essentially a random process. However, by treating the sequence of images as a set of samples from a multi... |

144 | Temporal texture modeling
- Szummer
- 1995
(Show Context)
Citation Context ...bitrarily in time: driving the model with random noise results in an infinitely varying sequence of images which always looks like the short input sequence. In this way, we can create “videotextures=-=” [21, 24] w-=-hich can play forever without repetition. With a moving camera, the image generation process comprises two components—a stochastic component generated by the videotexture, and a parametric component... |

111 | Automated Mosaicing with Super-resolution Zoom
- Capel, Zisserman
- 1998
(Show Context)
Citation Context ...age of a crowd applauding at a sports event. In these cases, it is not possible to impose the constraint that world points have similar colour in successive views, so existing registration techniques =-=[1, 5, 9, 11] c-=-annot be applied. Indeed the relationship between a point’s colours in successive frames is essentially a random process. However, by treating the sequence of images as a set of samples from a multi... |

74 |
Algorithms for the solution of the nonlinear least-squares problem
- Gill, Murray
- 1978
(Show Context)
Citation Context ...epanxx.mpg) shows significant misregistration, this is a good initial estimate. Then it is a simple matter to begin a nonlinear minimization. In this example, a modified Levenberg Marquardt algorithm =-=[8]-=- was used, with finite difference derivatives. A typical 6 iterations requires about 600 function evaluations. The minimizing parameters were used to create a stabilized sequence (in panxx2.mpg), whic... |

68 | The problem of degeneracy in structure and motion recovery from uncalibrated image sequences
- Torr, Fitzgibbon, et al.
- 1999
(Show Context)
Citation Context ... ��� � � � � ��� � � � � � �¨� ���� £ � § where the equality is up to scale. There are often considered to be two approaches to registration, “direct��=-=� and “feature-based” methods. References [14] and [26] a-=-re just two examples. This paper is most similar to direct methods, and indeed, if presented with a static scene, the algorithm � in 4 reduces to a (more expensive) direct correlation technique. As ... |

67 |
Independent Component Analysis: Principles and Practice
- Roberts, Everson
- 2001
(Show Context)
Citation Context ...e analysis tends to segment independent motions in the scene, choosing one or two components for each region that is moving. Of course, a more robust decomposition than PCA, perhaps K-PCA [22] or ICA =-=[19], -=-would enhance this effect and potentially improve performance. 2.2. AR model fitting Presenting the trimmed matrix of coefficients, X, to Neumaier and Schneider’sARfit algorithm [20], we obtain an e... |

65 | An assessment of information criteria for motion model selection
- Torr
- 1997
(Show Context)
Citation Context ...er and a family of maximum likelihood 3 estimators [23]. In this work, Yule-Walker estimators were used, although we plan to test MLE shortly. The choice of � can be made by model selection, e.g. AI=-=C [25]-=-. Recently, the work of Frey et al. [12, 7] has considered the problem of learning from unregistered sequences, and therefore solves a problem similar to that discussed here. An important difference b... |

63 | Estimating mixture models of images and inferring spatial transformations using the em algorithm
- Frey, Jojic
- 1999
(Show Context)
Citation Context ...estimators [23]. In this work, Yule-Walker estimators were used, although we plan to test MLE shortly. The choice of � can be made by model selection, e.g. AIC [25]. Recently, the work of Frey et al=-=. [12, 7]-=- has considered the problem of learning from unregistered sequences, and therefore solves a problem similar to that discussed here. An important difference between their work and ours is that they mod... |

62 | Representation of scenes from collections of images
- Kumar, Anandan, et al.
- 1995
(Show Context)
Citation Context ...� � ��� � ��� � � � � ��� � � � � � �¨� ���� £ � § where the equality is up to scale. There are often considered to be two approaches to regi=-=stration, “direct” and “feature-based” methods. References [14] a-=-nd [26] are just two examples. This paper is most similar to direct methods, and indeed, if presented with a static scene, the algorithm � in 4 reduces to a (more expensive) direct correlation techn... |

58 | A societry of models for video and image libraries
- Picard
- 1997
(Show Context)
Citation Context ...tion is based on deviation from causality of the scattergram of transformed intensities. 1.2. Time series analysis Computer vision has recently seen increased use of the tools of time series analysis =-=[2, 3, 13, 17, 18, 24], which model the temporal e-=-volution of physical systems. A time series is a sequence of vector-valued observations ��� � � ��� � � . One task of time series analysis is to forecast the value of � ���... |

43 | Transformed hidden Markov models: estimating mixture models of images and inferring spatial transformations in video sequences
- Jojic, Pretrovic, et al.
- 2000
(Show Context)
Citation Context ...estimators [23]. In this work, Yule-Walker estimators were used, although we plan to test MLE shortly. The choice of � can be made by model selection, e.g. AIC [25]. Recently, the work of Frey et al=-=. [12, 7]-=- has considered the problem of learning from unregistered sequences, and therefore solves a problem similar to that discussed here. An important difference between their work and ours is that they mod... |

26 |
Learning and recognising human dynamics in video sequences
- Bregler
- 1997
(Show Context)
Citation Context ...tion is based on deviation from causality of the scattergram of transformed intensities. 1.2. Time series analysis Computer vision has recently seen increased use of the tools of time series analysis =-=[2, 3, 13, 17, 18, 24], which model the temporal e-=-volution of physical systems. A time series is a sequence of vector-valued observations ��� � � ��� � � . One task of time series analysis is to forecast the value of � ���... |

26 | Classification of human body motion
- Rittscher, Blake
- 1999
(Show Context)
Citation Context |

23 | A Statistical Consistency Check for the space carving algorithm
- Broadhurst, Cipolla
- 2001
(Show Context)
Citation Context ...in successive layers are related by a compact time-series model. � so that the difference from a time series model. Replacing ����� the function here is similar in spirit to the work on sp=-=ace carving [4, 15]-=- where voxel consistency is measured as deviation from a lighting model. It also has the flavour of the early mutual information work [28] where the error metric for image registration is based on dev... |

22 |
Scene Matching by Hierarchical Correlation
- Glazer, Reynolds, et al.
- 1983
(Show Context)
Citation Context |

19 | A system for reconstruction of missing data in image sequences using sampled 3D AR models and MRF motion priors
- Kokaram, Godsill
- 1996
(Show Context)
Citation Context |

3 |
Algorithm: ARfit --- a Matlab package for estimation and spectral decomposition of multivariate autoregressive processes
- Schneider, Neumaier
- 1997
(Show Context)
Citation Context ...PCA [22] or ICA [19], would enhance this effect and potentially improve performance. 2.2. AR model fitting Presenting the trimmed matrix of coefficients, X, to Neumaier and Schneider’sARfit algorith=-=m [20], we obtain an e-=-stimate of the AR parameters �A ����� and the noise covari� ance �C. The model order may be chosen automatically using AIC, or input by the user. In the example in figure 3 the order w... |