#### DMCA

## Shape matching and object recognition using low distortion correspondence (2005)

### Cached

### Download Links

- [luthuli.cs.uiuc.edu]
- [luthuli.cs.uiuc.edu]
- [acberg.com]
- [www.cs.berkeley.edu]
- [acberg.com]
- [http.cs.berkeley.edu]
- [www.cs.berkeley.edu]
- [www.vis.uky.edu]
- [www.cnbc.cmu.edu]
- [www.eecs.berkeley.edu]
- [www.cs.cmu.edu]
- [www.cs.cmu.edu]
- [tamaraberg.com]
- [www.tamaraberg.com]
- [www.cise.ufl.edu]
- [www-2.cs.cmu.edu]
- [www.cs.cmu.edu]
- [www-2.cs.cmu.edu]
- [www.cs.berkeley.edu]
- [acberg.com]
- [www.tamaraberg.com]
- [tamaraberg.com]
- CiteULike

### Other Repositories/Bibliography

Venue: | In CVPR |

Citations: | 419 - 15 self |

### Citations

8954 | Distinctive image features from scale-invariant keypoint
- Lowe
- 2004
(Show Context)
Citation Context ...bout these points – they are not required to be keypoints such as those found using a Harris/Forstnerstype of operator or scale-space extrema of a Laplacian of Gaussian operator, such as used by Lowe =-=[18]-=-. We exploit three kinds of constraints to solve the correspondence problem between shapes: 1. Corresponding points on the two shapes should have similar local descriptors. There are several choices h... |

2739 | Object recognition from local scale-invariant features - Lowe - 1999 |

1783 | A performance evaluation of local descriptors,”
- Mikolajczyk, Schmid
- 2005
(Show Context)
Citation Context ...7], Schmid and Mohr [27], and Lowe [18] additionally use gray level information at the keypoints to provide greater discriminative power. Lowe’s SIFT descriptor has been shown in various studies e.g. =-=[22]-=- to perform very well particularly at tasks where one is looking for identical point features. Recent work extends this approach to category recognition [9, 7, 8], and to three-dimensional objects[26]... |

1184 | Robust real-time object detection.
- Viola, Jones
- 2002
(Show Context)
Citation Context ...without extracting either keypoints or edge points: Ullman et al propose using intermediate complexity features, a collection of image patches,[32]. For faces and cars the class specific detectors of =-=[33, 29, 28]-=- have been very successful. These techniques use simple local features, roughly based on image gradients, and a cascade of classifiers for efficiency. Recent work on sharing features [31] has extended... |

1127 | Object class recognition by unsupervised scale-invariant learning.
- Fergus, Perona, et al.
- 2003
(Show Context)
Citation Context ...tor has been shown in various studies e.g. [22] to perform very well particularly at tasks where one is looking for identical point features. Recent work extends this approach to category recognition =-=[9, 7, 8]-=-, and to three-dimensional objects[26]. It should be noted that not all objects have distinguished key points (think of a circle for instance), and using key points alone sacrifices the shape informat... |

784 | Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories.
- Fei-Fei, Fergus, et al.
- 2007
(Show Context)
Citation Context ...nd clutter. On average we achieve 48% accuracy on object classification with quite good localization on the correctly classified objects. This compares favorably with the state of the art of 16% from =-=[8]-=-. We also consider face detection for large faces, suitable for face recognition experiments. Here the task is to detect and localize a number of faces in an image. The face dataset we use is sampled ... |

658 | Comparing images using the Hausdorff distance
- Huttenlocher, Klanderman, et al.
- 1993
(Show Context)
Citation Context ... smooth portions of object contours. Approaches based on extracting edge points are, in our opinion, more universally applicable. Huttenlocher et al. developed methods based on the Hausdorff distance =-=[14]-=-. A drawback for our purposes is that the method does not return correspondences. Methods based on Distance Transforms, such as [11], are similar in spirit and behavior in practice. Work based on shap... |

637 | Distortion invariant object recognition in the dynamic linkarchitecture
- Lades, Vorbruggen, et al.
- 1993
(Show Context)
Citation Context ...ferent communities initiated such an approach: in computer vision, Fischler and Elschlager [10], in statistical image analysis, Grenander ( [12]and earlier), and in neural networks, von der Malsburg (=-=[15]-=- and earlier). The core idea that related but not identical shapes can be deformed into alignment using simple coordinate transformations dates even further back, to D’Arcy Thompson, in the 1910’s wit... |

625 | Learning to detect natural image boundaries using local brightness, color, and texture cues
- Martin, Fowlkes, et al.
(Show Context)
Citation Context ...ur of a signal at small number of locations {si}, indicated in (b.)sa. b. c. d. Figure 2. Two images (a. and c.) and four oriented edge channels derived from the images using the boundary detector of =-=[20]-=- (b. and d. respectively). A feature point descriptor is the concatenation of the subsampled geometric blur descriptor at the feature point for each of the channels. We use features based on a subsamp... |

548 | Local grayvalue invariants for image retrieval.
- Schmid, Mohr
- 1997
(Show Context)
Citation Context ...without explicitly solving for correspondences. Amit et al. [2] train decision trees for recognition by learning discriminative spatial configurations of keypoints. Leung et al. [17], Schmid and Mohr =-=[27]-=-, and Lowe [18] additionally use gray level information at the keypoints to provide greater discriminative power. Lowe’s SIFT descriptor has been shown in various studies e.g. [22] to perform very wel... |

464 |
A statistical method for 3D object detection applied to faces and cars,” in CVPR,
- Schneiderman, T
- 2000
(Show Context)
Citation Context ...without extracting either keypoints or edge points: Ullman et al propose using intermediate complexity features, a collection of image patches,[32]. For faces and cars the class specific detectors of =-=[33, 29, 28]-=- have been very successful. These techniques use simple local features, roughly based on image gradients, and a cascade of classifiers for efficiency. Recent work on sharing features [31] has extended... |

419 |
The Representation and Matching of Pictorial Structures.
- Fischler, Elschlager
- 1973
(Show Context)
Citation Context ...blem of deformable shape matching. Back in the 1970s, at least three different research groups working in different communities initiated such an approach: in computer vision, Fischler and Elschlager =-=[10]-=-, in statistical image analysis, Grenander ( [12]and earlier), and in neural networks, von der Malsburg ([15] and earlier). The core idea that related but not identical shapes can be deformed into ali... |

356 | A new point matching algorithm for non-rigid registration
- Chui, Rangarajan
(Show Context)
Citation Context ...ehavior in practice. Work based on shape contexts is indeed aimed at first finding correspondences [3, 23] and is close to the spirit of this work. Another approach is the non-rigid point matching of =-=[6]-=- based on thin plate splines and “softassign”. One can do without extracting either keypoints or edge points: Ullman et al propose using intermediate complexity features, a collection of image patches... |

309 | Sharing features: Efficient boosting procedures for multiclass object detection,” in
- Torralba, Murphy, et al.
- 2004
(Show Context)
Citation Context ...s of [33, 29, 28] have been very successful. These techniques use simple local features, roughly based on image gradients, and a cascade of classifiers for efficiency. Recent work on sharing features =-=[31]-=- has extended this to multiclass problems. 3. Geometric Blur Descriptor a. b. Figure 1. A sparse signal S (a.) and the geometric blur of S around the feature point marked in red (b.) We only sample th... |

245 |
Visual features of intermediate complexity and their use in classification
- Ullman, Vidal-Naquet, et al.
(Show Context)
Citation Context ...based on thin plate splines and “softassign”. One can do without extracting either keypoints or edge points: Ullman et al propose using intermediate complexity features, a collection of image patches,=-=[32]-=-. For faces and cars the class specific detectors of [33, 29, 28] have been very successful. These techniques use simple local features, roughly based on image gradients, and a cascade of classifiers ... |

243 | Real-time object detection for “smart” vehicles. Paper presented at the Computer Vision,
- Gavrila, Philomin
- 1999
(Show Context)
Citation Context ...ttenlocher et al. developed methods based on the Hausdorff distance [14]. A drawback for our purposes is that the method does not return correspondences. Methods based on Distance Transforms, such as =-=[11]-=-, are similar in spirit and behavior in practice. Work based on shape contexts is indeed aimed at first finding correspondences [3, 23] and is close to the spirit of this work. Another approach is the... |

211 | A Bayesian approach to unsupervised one-shot learning of object categories.
- Li, Fergus, et al.
- 2003
(Show Context)
Citation Context ...tor has been shown in various studies e.g. [22] to perform very well particularly at tasks where one is looking for identical point features. Recent work extends this approach to category recognition =-=[9, 7, 8]-=-, and to three-dimensional objects[26]. It should be noted that not all objects have distinguished key points (think of a circle for instance), and using key points alone sacrifices the shape informat... |

193 |
Feature detection in human vision: A phase-dependent energy model.
- Morrone, Burr
- 1988
(Show Context)
Citation Context ... one of two types of sparse channels from which to compute geometric blur descriptors: the oriented boundary detector output of [20] or oriented edge energy computed using quadrature pairs, following =-=[24]-=-. See Figure 2 for an example of the former. In each case the edge detector is used to produce four channels of oriented edge responses. Throughout we use a spatially varying Gaussian kernel to comput... |

170 | Matching Shapes,"
- Belongie, Malik, et al.
- 2001
(Show Context)
Citation Context ...straints to solve the correspondence problem between shapes: 1. Corresponding points on the two shapes should have similar local descriptors. There are several choices here: SIFT [18], Shape contexts =-=[3]-=-, and Geometric blur[4]. We use geometric blur. 2. Minimizing geometric distortion: If i and j are points on the model corresponding to i ′ and j ′ respectively, then the vector from i to j, �rij shou... |

149 | Geometric blur and template matching.
- AC, Malik
- 2001
(Show Context)
Citation Context ...orrespondence problem between shapes: 1. Corresponding points on the two shapes should have similar local descriptors. There are several choices here: SIFT [18], Shape contexts [3], and Geometric blur=-=[4]-=-. We use geometric blur. 2. Minimizing geometric distortion: If i and j are points on the model corresponding to i ′ and j ′ respectively, then the vector from i to j, �rij should be consistent with t... |

149 |
HANDS: A Pattern-Theoretic Study of Biological Shapes,
- Grenander, Chow, et al.
- 1991
(Show Context)
Citation Context ...70s, at least three different research groups working in different communities initiated such an approach: in computer vision, Fischler and Elschlager [10], in statistical image analysis, Grenander ( =-=[12]-=-and earlier), and in neural networks, von der Malsburg ([15] and earlier). The core idea that related but not identical shapes can be deformed into alignment using simple coordinate transformations da... |

138 | Object recognition with informative features and linear classification. - Vidal-Naquet, Ullman - 2003 |

135 |
Affine invariant model-based object recognition,
- Lamdan, Schwartz, et al.
- 1990
(Show Context)
Citation Context ...s this further in this paper. 2. Related Work There have been several approaches to shape recognition based on spatial configurations of a small number of keypoints or landmarks. In geometric hashing =-=[16]-=-, these configurations are used to vote for a model without explicitly solving for correspondences. Amit et al. [2] train decision trees for recognition by learning discriminative spatial configuratio... |

118 | Finding faces in cluttered scenes using random labeled graphmatching. In: ICCV.
- Leung, Burl, et al.
- 1995
(Show Context)
Citation Context ...d to vote for a model without explicitly solving for correspondences. Amit et al. [2] train decision trees for recognition by learning discriminative spatial configurations of keypoints. Leung et al. =-=[17]-=-, Schmid and Mohr [27], and Lowe [18] additionally use gray level information at the keypoints to provide greater discriminative power. Lowe’s SIFT descriptor has been shown in various studies e.g. [2... |

86 | Detecting and localizing edges composed of steps, peaks and roofs - Perona, Malik - 1990 |

84 | Joint induction of shape features and tree classifiers,
- Amit, Geman, et al.
- 1997
(Show Context)
Citation Context ...al configurations of a small number of keypoints or landmarks. In geometric hashing [16], these configurations are used to vote for a model without explicitly solving for correspondences. Amit et al. =-=[2]-=- train decision trees for recognition by learning discriminative spatial configurations of keypoints. Leung et al. [17], Schmid and Mohr [27], and Lowe [18] additionally use gray level information at ... |

84 | 3d object modeling and recognition using affine-invariant patches and multi-view spatial constraints,”
- Rothganger, Lazebnik, et al.
- 2003
(Show Context)
Citation Context ...[22] to perform very well particularly at tasks where one is looking for identical point features. Recent work extends this approach to category recognition [9, 7, 8], and to three-dimensional objects=-=[26]-=-. It should be noted that not all objects have distinguished key points (think of a circle for instance), and using key points alone sacrifices the shape information available in smooth portions of ob... |

83 | Shape contexts enable efficient retrieval of similar shapes” Computer Vision and Pattern Recognition,
- Mori, Belongie, et al.
- 2001
(Show Context)
Citation Context ...turn correspondences. Methods based on Distance Transforms, such as [11], are similar in spirit and behavior in practice. Work based on shape contexts is indeed aimed at first finding correspondences =-=[3, 23]-=- and is close to the spirit of this work. Another approach is the non-rigid point matching of [6] based on thin plate splines and “softassign”. One can do without extracting either keypoints or edge p... |

72 | A global solution to sparse correspondence problems,”
- Maciel, Costeira
- 2003
(Show Context)
Citation Context ...d the geometric distortion is encoded in a cost function defined over the space of correspondences. We purposely construct this to be an integer quadratic programming problem (cf. Maciel and Costeira =-=[19]-=-) and solve it using fast-approximate techniques. 1 We address two object recognition problems, multiclass recognition and face detection. In the multiple object class recognition problem, given an im... |

65 |
On growth and
- Thompson, d'Arcy
- 1966
(Show Context)
Citation Context ...e idea that related but not identical shapes can be deformed into alignment using simple coordinate transformations dates even further back, to D’Arcy Thompson, in the 1910’s with, On Growth and Form =-=[30]-=-. The basic subroutine in deformable matching takes as {aberg,millert,malik}@eecs.berkeley.edu input an image with an unknown object (shape) and compares it to a model by: solving the correspondence p... |

61 |
Feature-Centric Evaluation for Efficient Cascaded Object Detection.
- Schneiderman
- 2004
(Show Context)
Citation Context ...without extracting either keypoints or edge points: Ullman et al propose using intermediate complexity features, a collection of image patches,[32]. For faces and cars the class specific detectors of =-=[33, 29, 28]-=- have been very successful. These techniques use simple local features, roughly based on image gradients, and a cascade of classifiers for efficiency. Recent work on sharing features [31] has extended... |

53 |
Detection of local features invariant to affine transformations[D]. Institute National Polytechnique de
- Mikolajczyk
- 2002
(Show Context)
Citation Context ...0 exemplar faces our generic system provides a ROC curve with slightly better generalization, and slightly worse false detection rate than the quite effective specialized face detector of Mikolajczyk =-=[21]-=- used in [5]. 1 It is worth noting that this formulation is amenable to various probabilistic models, maximum likelihood estimation for a product of Gaussians among others, but we do not address this ... |

17 | Vidal-Naquet M and Sali E, Visual features of intermediate complexity and their use in classification. Nature Neuroscience - Ullman |

16 |
A thin plate spline method for mapping curves into curves in two dimensions
- Powell
- 1995
(Show Context)
Citation Context ...g Binary Quadratic Optimization to obtain a correspondence. Store the cost of the correspondence as well. 6. Extend the correpondence on m points to a smooth map using a regularized thin plate spline =-=[25]-=-. See Figures 3 and 7 for a number of examples. In the leftmost column of the figures is the image, P , shown with m points marked in color. In the middle left column is the target image Q with the co... |

9 |
Robust Real-time Object Detection,” 2nd international workshop on statistical and computational theories of vision –modeling, learning, computing, and sampling
- Viola, Jones
- 2001
(Show Context)
Citation Context |