## An affine invariant interest point detector (2002)

### Cached

### Download Links

Venue: | In Proceedings of the 7th European Conference on Computer Vision |

Citations: | 1063 - 43 self |

### BibTeX

@INPROCEEDINGS{Mikolajczyk02anaffine,

author = {Krystian Mikolajczyk and Cordelia Schmid},

title = {An affine invariant interest point detector},

booktitle = {In Proceedings of the 7th European Conference on Computer Vision},

year = {2002},

pages = {0--7}

}

### Years of Citing Articles

### OpenURL

### Abstract

Abstract. This paper presents a novel approach for detecting affine invariant interest points. Our method can deal with significant affine transformations including large scale changes. Such transformations introduce significant changes in the point location as well as in the scale and the shape of the neighbourhood of an interest point. Our approach allows to solve for these problems simultaneously. It is based on three key ideas: 1) The second moment matrix computed in a point can be used to normalize a region in an affine invariant way (skew and stretch). 2) The scale of the local structure is indicated by local extrema of normalized derivatives over scale. 3) An affine-adapted Harris detector determines the location of interest points. A multi-scale version of this detector is used for initialization. An iterative algorithm then modifies location, scale and neighbourhood of each point and converges to affine invariant points. For matching and recognition, the image is characterized by a set of affine invariant points; the affine transformation associated with each point allows the computation of an affine invariant descriptor which is also invariant to affine illumination changes. A quantitative comparison of our detector with existing ones shows a significant improvement in the presence of large affine deformations. Experimental results for wide baseline matching show an excellent performance in the presence of large perspective transformations including significant scale changes. Results for recognition are very good for a database with more than 5000 images.

### Citations

1881 | A Combined Corner and Edge Detector
- Harris, Stephens
(Show Context)
Citation Context ...ormation content in these points is high. One of the first recognition techniques based on interest points has been proposed by Schmid and Mohr [14]. The points are extracted with the Harris detector =-=[5]-=- which is invariant to image rotation. To obtain invariance to scale changes interest points can be extracted in the scale space of an image [7]. Dufournaud et al. [3] use a multi-scale framework to m... |

1770 | Object recognition from local scaleinvariant features - Lowe - 1999 |

911 | The Design and Use of Steerable Filters
- FREEMAN, ADELSON
- 1991
(Show Context)
Citation Context ...ves. Derivatives are computed on image patches normalized with the matrix U estimated for each point. Invariance to rotation is obtained by “steering” the derivatives in the direction of the gradient =-=[4]-=-. To obtain a stable estimation of the gradient direction, we use an average gradient orientation in a point neighbourhood. Invariance to affine intensity changes is obtained by dividing the derivativ... |

532 | Feature Detection with Automatic Scale Selection, in "International
- LINDEBERG
(Show Context)
Citation Context ...orithm for recognition based on local extrema of difference-of-Gaussian filters in scale-space. Mikolajczyk and Schmid [11] use a multi-scale framework to detect points and then apply scale selection =-=[8]-=- to select characteristic points. These points are invariant to scale changes and allow matching and recognition in the presence of large scale factors. Tuytelaars and Van Gool [16] detect affine inva... |

517 | Scale-space Theory in Computer Vision
- Lindeberg
- 1994
(Show Context)
Citation Context ...[14]. The points are extracted with the Harris detector [5] which is invariant to image rotation. To obtain invariance to scale changes interest points can be extracted in the scale space of an image =-=[7]-=-. Dufournaud et al. [3] use a multi-scale framework to match images at different scales. Interest points and descriptors are computed at several scales. A robust matching algorithm allows to select th... |

476 | Local Grayvalue Invariants for Image Retrieval
- Schmid, Mohr
- 1997
(Show Context)
Citation Context ...d reliably, are robust to partial visibility and the information content in these points is high. One of the first recognition techniques based on interest points has been proposed by Schmid and Mohr =-=[14]-=-. The points are extracted with the Harris detector [5] which is invariant to image rotation. To obtain invariance to scale changes interest points can be extracted in the scale space of an image [7].... |

339 | Indexing based on scale invariant interest points
- Mikolajczyk, Schmid
- 2001
(Show Context)
Citation Context ...th images in order to allow for correspondences. Features which have shown to be particularly appropriate are interest points. Scale invariant interest points detectors have been presented previously =-=[10, 11]-=-. However, none of the existing interest point detectors is invariant to affine transformations. In this paper wespresent an affine invariant interest point detector. For each interest point we simult... |

259 | Reliable feature matching across widely separated views
- Baumberg
- 2000
(Show Context)
Citation Context ...mpute an affine invariant Fourier description of the intensity profile along a line connecting two points. The description is not robust unless the two points lie on the same planar surface. Baumberg =-=[2]-=- extracts interest points at several scales and then adapts the shape of the region to the local image structure using an iterative procedure based on the second moment matrix [9]. Their descriptors a... |

177 | Wide baseline stereo matching based on local, affinely invariant regions
- Tuytelaars, Gool
- 2000
(Show Context)
Citation Context ...ply scale selection [8] to select characteristic points. These points are invariant to scale changes and allow matching and recognition in the presence of large scale factors. Tuytelaars and Van Gool =-=[16]-=- detect affine invariant regions based on image intensities. However, the number of such regions in an image is limited and depends on the content. They use colour descriptors computed for these regio... |

140 | Wide Baseline Stereo Matching - Pritchett, Zisserman - 1998 |

100 | Viewpoint invariant texture matching and wide baseline stereo
- Schaffalitzky, Zisserman
(Show Context)
Citation Context ... locations in scale space, converge to the same point location. The number of points is therefore reduced. The properties of the second moment matrix were also explored by Schaffalitzky and Zisserman =-=[13]-=-, but their goal was to obtain an affine invariant texture descriptor. 1.2 Our approach A uniform Gaussian scale-space is often used to deal with scale changes [3, 7, 10, 11]. However, an affine Gauss... |

63 | Matching Image with Different Resolutions
- Dufournaud, Schmid, et al.
(Show Context)
Citation Context ...tracted with the Harris detector [5] which is invariant to image rotation. To obtain invariance to scale changes interest points can be extracted in the scale space of an image [7]. Dufournaud et al. =-=[3]-=- use a multi-scale framework to match images at different scales. Interest points and descriptors are computed at several scales. A robust matching algorithm allows to select the correct scale. In the... |

57 |
Geometric information criterion for model selection
- Kanatani
- 1998
(Show Context)
Citation Context ...ased on RANdom SAmple Consensus (RANSAC) rejects inconsistent matches. For our experimental results the transformation used is either a homography or a fundamental matrix. A model selection algorithm =-=[6]-=- can be used to automatically decide which transformation is the most appropriate one. Database retrieval. A voting algorithm is used to select the most similar images in the database. This makes retr... |

57 |
Wide baseline point matching using affine invariants computed from intensity profiles
- Tell, Carlsson
- 2000
(Show Context)
Citation Context ...mographies are obtained by matching regions bound by four line segments. This approach has been applied to wide baseline matching and it is clearly difficult to extend to retrieval. Tell and Carlsson =-=[15]-=- also address the problem of wide baseline matching and use an affine invariant descriptors for point pairs. They compute an affine invariant Fourier description of the intensity profile along a line ... |

56 | Shape-adapted smoothing in estimation of 3-d shape cues from affine deformations of local 2-d brightness structure
- Lindeberg, Garding
- 1997
(Show Context)
Citation Context ...ar surface. Baumberg [2] extracts interest points at several scales and then adapts the shape of the region to the local image structure using an iterative procedure based on the second moment matrix =-=[9]-=-. Their descriptors are affine invariant for fixed scale and location, that is the scale and the location of the points are not extracted in an affine invariant way. The points as well as the associat... |

41 | Lindederg T., Fingerprint Enhancement by Shape Adaptation of Scale-space Operators with Automatic Scale Selection
- Almansa
(Show Context)
Citation Context ...scale allows to obtain a reasonable eigenvalue ratio and allows convergence for pointsswhich would not converge if the ratio is too large. A similar approach for local scale selection was proposed in =-=[1]-=-. Spatial localization. It is well known that the local maxima of the Harris measure (equation 2) change their spatial location if the detection scale changes. This can also be observed if the scale c... |

4 | Viewpoint invariant texture matching and wide baseline stereo - Schaalitzky, Zisserman - 2001 |

2 | Matching images with di erent resolutions - Dufournaud, Schmid, et al. - 2000 |

2 | Wide baseline point matching using a#ne invariants computed from intensity profiles - Tell, Carlsson - 2000 |