## Segmentation using eigenvectors: A unifying view (1999)

### Cached

### Download Links

Venue: | In ICCV |

Citations: | 334 - 1 self |

### BibTeX

@INPROCEEDINGS{Weiss99segmentationusing,

author = {Yair Weiss},

title = {Segmentation using eigenvectors: A unifying view},

booktitle = {In ICCV},

year = {1999}

}

### Years of Citing Articles

### OpenURL

### Abstract

Automatic grouping and segmentation of images remains a challenging problem in computer vision. Recently, a number of authors have demonstrated good performance on this task using methods that are based on eigenvectors of the a nity matrix. These approaches are extremely attractive in that they are based on simple eigendecomposition algorithms whose stability is well understood. Nevertheless, the use of eigendecompositions in the context of segmentation is far from well understood. In this paper we give a unied treatment of these algorithms, and show the close connections between them while highlighting their distinguishing features. We then prove results on eigenvectors of block matrices that allow us to analyze the performance of these algorithms in simple grouping settings. Finally, we use our analysis to motivate a variation on the existing methods that combines aspects from di erent eigenvector segmentation algorithms. We illustrate our analysis with results on real and synthetic images. Human perceiving a scene can often easily segment it into coherent segments or groups. There has been a tremendous amount of e ort devoted to achieving the same level of performance in computer vision. In many cases, this is done by associating with each pixel a feature vector (e.g. color, motion, texture, position) and using a clustering or grouping algorithm on these feature vectors. Perhaps the cleanest approach to segmenting points in feature space is based on mixture models in which one assumes the data were generated by multiple processes and estimates the parameters of the processes and the number of components in the mixture. The assignment of points to clusters can then be easily performed by calculating the posterior probability ofa point belonging to a cluster. Despite the elegance of this approach, the estimation process leads to a notoriously di cult optimization. The frequently used EM algorithm [3] often converges to a local maximum that depends on the initial conditions. Recently, anumber of authors [11, 10, 8, 9, 2] have suggested alternative segmentation methods that are based on eigenvectors of the (possibly normalized) \a nity matrix". Figure 1a shows two clusters of points and gure 1b shows the a nity matrix de ned by:

### Citations

9033 | Maximum likelihood from incomplete data via the EM algorithm
- Dempster, Laird, et al.
- 1977
(Show Context)
Citation Context ... posterior probability of a point belonging to a cluster. Despite the elegance of this approach, the estimation process leads to a notoriously difficult optimization. The frequently used EM algorithm =-=[3]-=- often converges to a local maximum that depends on the initial conditions. Recently, a number of authors [11, 10, 8, 9, 2] have suggested alternative segmentation methods that are based on eigenvecto... |

2774 | Normalized cuts and image segmentation
- Shi, Malik
- 2000
(Show Context)
Citation Context ...on process leads to a notoriously difficult optimization. The frequently used EM algorithm [3] often converges to a local maximum that depends on the initial conditions. Recently, a number of authors =-=[11, 10, 8, 9, 2] have sugg-=-ested alternative segmentation methods that are based on eigenvectors of the (possibly normalized) "affinity matrix". Figure 1a shows two clusters of points and figure 1b shows the affinity ... |

1048 |
Spectral Graph Theory
- Chung
- 1997
(Show Context)
Citation Context ... approaches appear quite different. There is, however, a closer connection. Define the normalized affinity matrix: N = D \Gamma1=2 WD \Gamma1=2 (6) We call this a normalized affinity matrix following =-=[1]-=-. Note that N (i; j) = W (i; j)= p D(i; i) p D(j; j). Given N the following normalization lemma is easily shown: Normalization Lemma: 1. Let v be an eigenvector of N with eigenvaluesthen D \Gamma1=2 v... |

154 | A factorization approach to grouping
- Perona, Freeman
- 1998
(Show Context)
Citation Context ...on process leads to a notoriously difficult optimization. The frequently used EM algorithm [3] often converges to a local maximum that depends on the initial conditions. Recently, a number of authors =-=[11, 10, 8, 9, 2] have sugg-=-ested alternative segmentation methods that are based on eigenvectors of the (possibly normalized) "affinity matrix". Figure 1a shows two clusters of points and figure 1b shows the affinity ... |

148 | A multibody factorization method for motion analysis
- Costeira, Kanade
- 1995
(Show Context)
Citation Context ...on process leads to a notoriously difficult optimization. The frequently used EM algorithm [3] often converges to a local maximum that depends on the initial conditions. Recently, a number of authors =-=[11, 10, 8, 9, 2] have sugg-=-ested alternative segmentation methods that are based on eigenvectors of the (possibly normalized) "affinity matrix". Figure 1a shows two clusters of points and figure 1b shows the affinity ... |

91 |
Quantitative Measures of Change Based on Feature Organization
- Sarkar, Boyer
- 1996
(Show Context)
Citation Context |

63 |
Multibody grouping from motion images
- Gear
- 1998
(Show Context)
Citation Context ...dy rigid grouping, there has been additional progress made by using algorithms that do not use eigendecompositions but rather other, more stable matrix decompositions such as the reduced echelon form =-=[4, 5]-=-. Given the close connection between the two problems, we are currently experimenting with using these alternative decompositions in the general grouping context. The main goal of presenting these alg... |

41 | Feature grouping by â€™relocalisationâ€™ of eigenvectors of the proximity matrix
- Scott
- 1990
(Show Context)
Citation Context |

19 | Self inducing relational distance and its application to image segmentation
- Shi, Malik
- 1998
(Show Context)
Citation Context ...mbines the SM algorithm with the SLH algorithm --- work with eigenvectors of normalized W but use the first k eigenvectors rather than just the first two. This is similar in spirit to the approach of =-=[12]-=- where the first k eigenvectors of W were used to define a new affinity matrix between the points. Our experimental results on real images are encouraging --- by using the first k eigenvectors and com... |

13 | Class representation and image retrieval with non-metric distances
- Jacobs, Weinshall, et al.
- 2000
(Show Context)
Citation Context ...oe a free parameter. In this case we have used d(x i ; x j ) = kx i \Gamma x j k 2 but different definition of affinities are possible. The affinities do not even have to obey the metric axioms (e.g. =-=[7]-=-), we will only assume that d(x i ; x j ) = d(x j ; x i ). Note that we have ordered the points so that all points belonging to the first cluster appear first and the points in the second cluster. Thi... |

2 |
feature grouping in moving images
- Gear
- 1994
(Show Context)
Citation Context ...eigenvectors of affinity matrices to segment points in feature space, there has been interest in using singular values of the measurement matrix to segment the points into rigidly moving bodies in 3D =-=[2, 4]-=-. Although these algorithms seem quite different from the ones discussed so far, they are in fact very closely related. To see the connection, we review the Costeira and Kanade algorithm. Suppose we t... |