## Recognition without Correspondence using Multidimensional Receptive Field Histograms (2000)

### Cached

### Download Links

- [www.vision.ethz.ch]
- [www.cs.ubc.ca]
- [whitechapel.media.mit.edu]
- DBLP

### Other Repositories/Bibliography

Venue: | International Journal of Computer Vision |

Citations: | 216 - 19 self |

### BibTeX

@ARTICLE{Schiele00recognitionwithout,

author = {Bernt Schiele and James L. Crowley},

title = {Recognition without Correspondence using Multidimensional Receptive Field Histograms},

journal = {International Journal of Computer Vision},

year = {2000},

volume = {36},

pages = {31--50}

}

### Years of Citing Articles

### OpenURL

### Abstract

. The appearance of an object is composed of local structure. This local structure can be described and characterized by a vector of local features measured by local operators such as Gaussian derivatives or Gabor filters. This article presents a technique where appearances of objects are represented by the joint statistics of such local neighborhood operators. As such, this represents a new class of appearance based techniques for computer vision. Based on joint statistics, the paper develops techniques for the identification of multiple objects at arbitrary positions and orientations in a cluttered scene. Experiments show that these techniques can identify over 100 objects in the presence of major occlusions. Most remarkably, the techniques have low complexity and therefore run in real-time. 1. Introduction The paper proposes a framework for the statistical representation of the appearance of arbitrary 3D objects. This representation consists of a probability density function or jo...

### Citations

2910 | Eigenfaces for recognition - Turk, Pentland - 1991 |

2779 |
Introduction to Statistical Pattern Recognition, 2nd ed
- Fukunaga
- 1990
(Show Context)
Citation Context ...ecently many researchers (Sirovich and Kirby, 1987; Turk and Pentland, 1991; Murase and Nayar, 1995; Moghaddam and Pentland, 1995; Ohba and Ikeuchi, 1996) have used the Karhunen-Loeve transformation (=-=Fukunaga, 1990-=-) for the calculation of eigenpictures in the context of object recognition. The main advantage of this approach is the representation of each image by a small number of coefficients, which can be sto... |

1377 | Color indexing - Swain, Ballard - 1991 |

984 | Visual learning and recognition of 3-D objects from appearance - Murase, Nayar - 1995 |

918 |
Query by image and video content: The QBIC system
- Flickner, Sawhney, et al.
- 1995
(Show Context)
Citation Context .... The color histogram approach is an attractive method for object recognition, because of its simplicity, speed and robustness. Many image retrieval system use color histograms among other cues (e.g. =-=Flickner et al., 1995-=-; Belongie et al., 1998) which is motivated by the fact that many images contain characteristic colors. Since many objects cannot be described by color alone thissRecognition without Correspondence us... |

879 |
Theory of communication
- Gabor
- 1946
(Show Context)
Citation Context ...ell assRecognition without Correspondence using Multidimensional Receptive Field Histograms 35 Gabor filters) are robust to scale changes of approximately ±20% (Schmid and Mohr, 1997). Gabor filters (=-=Gabor, 1946-=-; Westelius, 1992; Daugman, 1993) satisfy the same constraints as Gaussian filters (robustness, steerability to image plane rotation, equivariance to scale changes). During earlier experiments (not re... |

878 | The design and use of steerable filters
- Freeman, Adelson
- 1991
(Show Context)
Citation Context ... of other objects. More specifically, the feature vector includes 45 responses of nine oriented Gaussian filters at five different scales (9 × 5 = 45). Using the steerability of Gaussian derivatives (=-=Freeman and Adelson, 1991-=-), the feature vector is made rotational invariant. During training and object recognition a figure ground segmentation is performed and the vectors are stored in a generalized version of Karneva’s sp... |

570 |
Statistical and structural approaches to texture
- Haralick
- 1979
(Show Context)
Citation Context ...ors (Nagao, 1995; Funt and Finlayson, 1995) provide a natural extension of the proposed statistical object representation technique described below. One can also consider the use of texture features (=-=Haralick, 1979-=-; Mao and Jain, 1992) or low–level geometric features and perceptual significant groups thereof (Pope and Lowe, 1996). 3.1. Gaussian Derivatives Gaussian derivatives are widely used in the literature ... |

469 | Local Grayvalue Invariants for Image Retrieval - Schmid, Mohr - 1997 |

452 | Low-dimensional procedure for the characterization of human faces - Sirovich, Kirby - 1987 |

431 |
Confidence Visual Recognition of Persons by a Test of Statistical Independence
- Daugman, “High
- 1993
(Show Context)
Citation Context ...respondence using Multidimensional Receptive Field Histograms 35 Gabor filters) are robust to scale changes of approximately ±20% (Schmid and Mohr, 1997). Gabor filters (Gabor, 1946; Westelius, 1992; =-=Daugman, 1993-=-) satisfy the same constraints as Gaussian filters (robustness, steerability to image plane rotation, equivariance to scale changes). During earlier experiments (not reported below) Gabor filters obta... |

334 |
Geometric hashing: a general and efficient Model-Based Recognition Scheme
- Lamdan, Wolfson
- 1988
(Show Context)
Citation Context ...orrespondence using Multidimensional Receptive Field Histograms 49 2. In the case of a projective transformation a four point basis is used. For similarity transformations only two points are needed (=-=Lamdan and Wolfson, 1988-=-). 3. The responses of the 45-dimensional feature vector are calculated with a 8 × 8 kernel at five different levels of an image pyramid. Since each level of the image pyramid is reduced by a factor o... |

263 |
Efficient Color Histogram Indexing for Quadratic Form Distance Function
- Hafner, Sawhney, et al.
- 1995
(Show Context)
Citation Context ...er recognition results for most cases than the intersection measurement with respect to image distortions due to appearance changes, additive Gaussian noise and blur. Even though quadratic distances (=-=Hafner et al., 1995-=-) were typically outperformed by intersection and χ 2 , the Mahalanobis distance—as a special case of quadratic distances—sometimes obtains comparable results (Schiele, 1997). Object recognition by me... |

258 | Representation of local geometry in the visual system - Koenderink, Doorn - 1987 |

242 |
G.D.: Color constant color indexing
- Funt, Finlayson
- 1995
(Show Context)
Citation Context ...riments (not reported below) Gabor filters obtained almost identical results as Gaussian derivatives. Even though color has not been used in our experiments, invariant color descriptors (Nagao, 1995; =-=Funt and Finlayson, 1995-=-) provide a natural extension of the proposed statistical object representation technique described below. One can also consider the use of texture features (Haralick, 1979; Mao and Jain, 1992) or low... |

241 | Using Canny’s criteria to derive a recursively implemented optimal edge detector. Int J Comput Vision 1:167–187
- Deriche
- 1987
(Show Context)
Citation Context ...umbers of image patches resemble derivatives of Gaussians (Rao and Ballard, 1995)), their capacity to model the response of neural cells (Young, 1986) and the existence of a recursive implementation (=-=Deriche, 1987-=-). Furthermore, Gaussian derivatives (as well assRecognition without Correspondence using Multidimensional Receptive Field Histograms 35 Gabor filters) are robust to scale changes of approximately ±20... |

237 |
Geometric invariance in computer vision
- Mundy, Zisserman
- 1992
(Show Context)
Citation Context ...the number of free parameters is to choose local image measurements which are invariant to different parameters. Such invariant properties are used by many researchers (Burkhardt and Zisserman, 1992; =-=Mundy and Zisserman, 1992-=-; Mundy et al., 1993) and applied successfully in various ways. Unfortunately the obtained invariants are very restrictive to certain types of objects. Robust or quasi-invariant local image measuremen... |

218 | Object Recognition using Multidimensional Receptive Field Histograms - Schiele - 1997 |

195 |
Texture classification and segmentation using multiresolution simultaneous autoregressive models
- Mao, Jain
- 1992
(Show Context)
Citation Context ...; Funt and Finlayson, 1995) provide a natural extension of the proposed statistical object representation technique described below. One can also consider the use of texture features (Haralick, 1979; =-=Mao and Jain, 1992-=-) or low–level geometric features and perceptual significant groups thereof (Pope and Lowe, 1996). 3.1. Gaussian Derivatives Gaussian derivatives are widely used in the literature and well understood ... |

131 | Deformable kernels for early vision - Perona - 1995 |

131 | An active vision architecture based on iconic representations - Rao, Ballard - 1995 |

119 | Computational framework for determining stereo correspondence from a set of linear spatial filters - Jones, Malik - 1992 |

118 | Object recognition by affine invariant matching - LAMDAN, SCHWARTZ, et al. - 1988 |

91 | Comparing and evaluating interest points - Schmid, Mohr, et al. - 1998 |

87 | Dynamic model of visual recognition predicts neural response properties in the visual cortex - Rao, Ballard - 1997 |

85 | Recursively implementing the Gaussian and its derivatives - Deriche - 1992 |

80 |
Colorand texture-based image segmentation using EM and its application to content-based image retrieval
- Belongie, Carson, et al.
- 1998
(Show Context)
Citation Context ...pproach is an attractive method for object recognition, because of its simplicity, speed and robustness. Many image retrieval system use color histograms among other cues (e.g. Flickner et al., 1995; =-=Belongie et al., 1998-=-) which is motivated by the fact that many images contain characteristic colors. Since many objects cannot be described by color alone thissRecognition without Correspondence using Multidimensional Re... |

68 |
Space and time bounds on indexing 3D models from 2D images
- Clemens, Jacobs
- 1991
(Show Context)
Citation Context ...tion is managed by the rotational steerability of local descriptors. The two rotational degrees of freedom rx and ry represent a viewpoint change of the observer. Several authors (Burns et al., 1990; =-=Clemens and Jacobs, 1991-=-) show the non-existence of viewpoint invariant descriptors for the general case. Nevertheless, useful descriptors exist in special cases (Mundy and Zisserman, 1992; Mundy et al., 1993). As mentioned ... |

60 | Object indexing using an iconic sparse distributed memory - Rao, Ballard - 1995 |

53 |
Finding Waldo, or Focus of Attention Using Local Color Information
- Ennesser, Medioni
- 1995
(Show Context)
Citation Context ...ally constant illumination). More recently Finlayson et al. (1998) introduced a color image normalization which is invariant to light intensity and light color changes. Another interesting extension (=-=Ennesser and Medioni, 1995-=-) uses local color histograms of the test image in order to deal with more cluttered scenes. Since not all objects can be described and recognized by color alone, color histograms have been combined w... |

51 | Comprehensive colour image normalization - Finlayson, Schiele, et al. - 1998 |

49 | Model-Based Object Recognition by Geometric Hashing - Wolfson - 1990 |

46 | Affine Matching with Bounded Sensor Error: A Study of Geometric Hashing and Alignment - Grimson, Huttenlocher, et al. - 1991 |

37 | Maximum likelihood detection of faces and hands - Moghaddam, Pentland - 1995 |

34 |
View variation of point-set and line segment features
- Burns, Weiss, et al.
- 1990
(Show Context)
Citation Context ...rs, image plane rotation is managed by the rotational steerability of local descriptors. The two rotational degrees of freedom rx and ry represent a viewpoint change of the observer. Several authors (=-=Burns et al., 1990-=-; Clemens and Jacobs, 1991) show the non-existence of viewpoint invariant descriptors for the general case. Nevertheless, useful descriptors exist in special cases (Mundy and Zisserman, 1992; Mundy et... |

26 | The Complexity of Perceptual Search Tasks - Tsotsos - 1989 |

24 | Object recognition using steerable filters at multiple scales - Ballard, Wixson - 1993 |

19 | Distributed Bayesian object recognition - Rigoutsos, Hummel - 1993 |

17 | Statistical learning, localization and identification of objects - Hornegger, Niemann - 1995 |

17 | ÒSimulation of Human Retinal Function with the Gaussian Derivative ModelÓ - Young - 1986 |

16 | Using illumination invariant color histogram descriptors for recognition - Healey, Slater - 1994 |

15 | Three-dimensional object recognition using an unsupervised neural network Understanding the distinguishing features
- Intrator, Gold, et al.
- 1991
(Show Context)
Citation Context ...ate of such a high-dimensional probability density function. The difficulty is due to the fact that the number of training examples is exponential in the number of dimensions of the density function (=-=Intrator and Gold, 1993-=-). The most effective way to reduce the number of free parameters is to choose local image measurements which are invariant to different parameters. Such invariant properties are used by many research... |

15 | A computational model of texture segmentation - Malik, Perona - 1989 |

14 | Combining color and geometric information for the illuminantion invariant recognition - Slater, Healey - 1995 |

13 | Bayesian decision versus voting for image retrieval
- Mohr, Picard, et al.
- 1997
(Show Context)
Citation Context ...s made more selective by combining the vector responses with geometrical invariants between different interest points. Another possibility for improvement is the use of a probabilistic voting scheme (=-=Mohr et al., 1997-=-). Impressive experimental results have been presented on a database of several hundred objects. Nevertheless, arguably the weakest point of the approach iss34 Schiele and Crowley the application of a... |

12 | Learning to Recognize Objects in Images: Acquiring and Using Probabilistic Models of Appearance - Pope - 1995 |

12 | Preattentive gaze control for robot vision - Westelius - 1992 |

11 | Recognizing 3D objects using photometric invariant - Nagao - 1995 |

10 | Cluster-based probability model applied to image restoration and compression - Popat, Picard - 1994 |

8 | Combining grayvalue invariants with local constraints for object recognition - Schmid, Mohr - 2002 |