## Model-Based Object Recognition - A Survey of Recent Research (1994)

Citations: | 65 - 1 self |

### BibTeX

@TECHREPORT{Pope94model-basedobject,

author = {Arthur R. Pope},

title = {Model-Based Object Recognition - A Survey of Recent Research},

institution = {},

year = {1994}

}

### Years of Citing Articles

### OpenURL

### Abstract

We survey the main ideas behind recent research in model-based object recognition. The survey covers representations for models and images and the methods used to match them. Perceptual organization, the use of invariants, indexing schemes, and match verification are also reviewed. We conclude that there is still much room for improvement in the scope, robustness, and efficiency of object recognition methods. We identify what we believe are the ways improvements will be achieved. ii Contents 1. Introduction .................................................................................................................................... 1 2. Representation ................................................................................................................................ 3 2.1 What makes a good shape representation? ............................................................................ 3 2.2 The choice of coordinate system ..........................................

### Citations

1035 |
Face Recognition Using Eigenfaces
- Turk, Pentland
- 1991
(Show Context)
Citation Context ...ly prevalent among the objects to be recognized. Others have used similar unsupervised learning techniques to automatically develop features that are 2D patterns of intensity or its derivative (e.g., =-=Turk and Pentland 1991-=-; Murase and Nayar 1993; Weng, Ahuja, and Huang 1993). Thus far we have considered what types of shape primitives a representation might employ. Also important is the issue of what sizes or scales tho... |

495 |
Representation and recognition of the spatial organization of three-dimensional shapes
- Marr, Nishihara
- 1978
(Show Context)
Citation Context ...on? Many researchers have prefaced their proposals for shape representations with a discussion of the criteria that ought to be satisfied by such a representation. These may be found, for example, in =-=Marr and Nishihara 1978-=-; Binford 1982; Brady 1983; Woodham 1987; Haralick, Mackworth, and Tanimoto 1988; and Mokhtarian and Mackworth 1992. The following have been mentioned frequently in some manner: 4 scope and sensitivit... |

473 | Some informational aspects of visual perception - Attneave - 1954 |

472 |
Perceptual Organization and Visual Recognition
- Lowe
- 1985
(Show Context)
Citation Context ...rtain relations between lines in three dimensions, such as cotermination (the proximity of endpoints), parallelism, and collinearity, appear approximately invariant when seen from various viewpoints (=-=Lowe 1985-=-). Even 8 angles between general pairs of lines (not necessarily parallel ones), and ratios of line lengths, may be relatively stable with respect to viewpoint (Ben-Arie 1990; Burns, Weiss, and Risema... |

428 | Recognition by Linear Combinations of Models - Ullman, Basri - 1991 |

357 |
Object Recognition by Computer: The Role of Geometric Constraints
- Grimson
- 1990
(Show Context)
Citation Context ...ts whose shapes may differ in certain ways. One approach is to use a parameterized model in which free variables or quantifiers are used to specify certain measurements (e.g., Brooks 1981; Lowe 1989; =-=Grimson 1990-=-). A model for pencils, for example, might describe the pencil body with a generalized cylinder whose length is a free variable; one for scissors might use a free variable to describe the rotation of ... |

308 |
A network that learns to recognize three-dimensional objects
- Poggio, Edelman
- 1990
(Show Context)
Citation Context ...dimensions. A markedly different approach to indexing has been developed using networks of neuron-like units that compute functions called generalized radial basis functions or hyper-basis functions (=-=Poggio and Edelman 1990-=-; Brunelli and Poggio 1991; Edelman and Poggio 1990). A modeled object is represented by a network in which individual units represent distinct prototypical or characteristic views of the object. The ... |

262 |
S.: Recognizing solid objects by alignment with an image
- HUTTENLOCHER, ULLMAN
- 1990
(Show Context)
Citation Context ...Turney, and Mudge 1989; Lamdan, Schwartz, and Wolfson 1990). Some also assess negative evidence, such as image edges that cross projected model edges at large angles (e.g., Hansen and Henderson 1989; =-=Huttenlocher and Ullman 1990-=-). Match solutions are verified by testing these measures against empirically determined thresholds, and then ranked according to the measures to select the best, mutually-consistent solutions. Grimso... |

239 |
Geometric Invariance in Computer Vision
- Mundy, Zisserman
- 1992
(Show Context)
Citation Context ...where the effects of lighting and pose are confounded. Recently, considerable effort has been directed at identifying and employing primitives that are completely invariant with respect to viewpoint (=-=Mundy and Zisserman 1992-=-; Weiss 1993). These are based on properties of geometric structures, called invariants or geometric invariants, that remain constant over an entire class of transformations. Each is defined in terms ... |

219 |
A theory of multiscale, curvature-based shape representation for planar curves
- Mokhtarian, Mackworth
- 1992
(Show Context)
Citation Context ...a that ought to be satisfied by such a representation. These may be found, for example, in Marr and Nishihara 1978; Binford 1982; Brady 1983; Woodham 1987; Haralick, Mackworth, and Tanimoto 1988; and =-=Mokhtarian and Mackworth 1992-=-. The following have been mentioned frequently in some manner: 4 scope and sensitivity The representation must be able to describe all relevant shapes while preserving all important distinctions among... |

200 | The Curvature Primal Sketch
- Asada, Brady
- 1986
(Show Context)
Citation Context ...lication in mind so that, although they may not describe a shape completely, they will capture the information that is important. The curvature primal sketch (CPS) is 12 one example of this approach (=-=Asada and Brady 1986-=-). It represents 2D curves only in terms of the significant changes of curvature---including corners and inflections---that are detected at each scale. Among the motivations for the CPS is a hypothesi... |

197 | Symbolic reasoning among 3-D models and 2-D images - Brooks - 1981 |

183 |
Structural descriptions and inexact matching
- Shapiro, Haralick
- 1981
(Show Context)
Citation Context ...st score yet obtained for a complete interpretation. Correspondence space search has often been cast as a problem of graph matching (e.g., Ben-Arie and Meiri 1987; Bergevin and Levine 1993; Fan 1990; =-=Shapiro and Haralick 1981-=-; Wong 1992; Yang, Snyder, and Bilbro 1989; Zhang, Sullivan, and Baker 1992). In this framework, the task is to find a common subgraph isomorphism between two attributed graphs: one representing the i... |

181 |
Aligning Pictorial Descriptions: An Approach to Object Recognition
- Ullman
- 1989
(Show Context)
Citation Context ...1992b), Burns and Riseman (1992), and Camps, Shapiro, and Haralick (1991). There is some interesting evidence that the human visual system uses a viewer-centred representation for object recognition (=-=Ullman 1989-=-; Edelman and Blthoff 1992). Humans are able to recognize objects more accurately and rapidly when they are seen from particular viewpoints, implying that those views of an object are readily availabl... |

145 | 3-d shape recovery using distributed aspect matching - Dickinson, Pentland, et al. - 1992 |

137 |
Invariant descriptors for 3d object recognition and pose
- Forsyth, Mundy, et al.
- 1991
(Show Context)
Citation Context ...1988). . A pair of coplanar conics yields two projective invariants. These can be used to identify objects like gaskets, which are essentially coplanar and typically composed of circles and ellipses (=-=Forsyth et al. 1991-=-). . A rotationally symmetric object has a silhouette that is essentially planar. Therefore invariants of coplanar features can be used to identify a rotationally symmetric object by its silhouette (F... |

131 | HYPER: A new approach for the recognition and positioning of twodimensional objecs
- Ayache, Faugeras
- 1986
(Show Context)
Citation Context ... to yield stable descriptions of some shapes. Many systems have been demonstrated that use segments or patches as shape primitives, including HYPER , which uses polygonal approximations of 2D curves (=-=Ayache and Faugeras 1986-=-), and 3D-POLY, which uses quadric approx - imations of 3D surfaces (Chen and Kak 1989). b) Parts With a limited set of parts one can construct a large variety of objects, especially if each part can ... |

121 | Object recognition by affine invariant matching - Lamdan, Schwartz, et al. - 1988 |

120 |
Recognizing and locating partially visible objects: The localfeature-focus method
- Bolles, Cain
- 1982
(Show Context)
Citation Context ...considers the likelihood of features arising accidentally, the cost of detecting them, and the degree to which they may be distorted by perspective. Another variation, the local feature focus method (=-=Bolles and Cain 1982-=-), involves analyzing the entire database of object models to select for each model one or more focus features that, due to their specificity, can be used to initiate matching by alignment. There has ... |

115 | On the sensitivity of the hough transform for object recognition
- Grimson, Huttenlocher
- 1990
(Show Context)
Citation Context ..., when the image contains a great deal of clutter, random clusters of votes may overshadow a correct solution so that a large portion of the array must be examined before a correct solution is found (=-=Grimson and Huttenlocher 1990-=-). Cass (1992) and Breuel (1992a) have developed transformation space algorithms that avoid the problems that are due to tessellation. Like the generalized Hough transform, they use a bounded error mo... |

114 | Orientation dependence in the recognition of familiar and novel views of threedimensional objects
- Edelman, Bülthoff
- 1992
(Show Context)
Citation Context ... and Riseman (1992), and Camps, Shapiro, and Haralick (1991). There is some interesting evidence that the human visual system uses a viewer-centred representation for object recognition (Ullman 1989; =-=Edelman and Blthoff 1992-=-). Humans are able to recognize objects more accurately and rapidly when they are seen from particular viewpoints, implying that those views of an object are readily available while others must be com... |

94 | Face recognition based on depth and curvature features - Gordon - 1991 |

82 | Computing exact aspect graphs of curved objects: algebraic surfaces - Petitjean, Kriegman, et al. - 1992 |

74 |
Perceptual organization for scene segmentation and description
- Mohan, Nevatia
- 1992
(Show Context)
Citation Context ...t principles. Instead, it has been common practice to develop ad hoc criteria for deciding which sets of features should be grouped (e.g., Bergevin and Levine 1992; Horaud, Veillon, and Skordas 1990; =-=Mohan and Nevatia 1992-=-; Sarkar and Boyer 1990; Saund 1990; Stein and Medioni 1992a). Researchers concerned with the problem of how to identify groups efficiently have produced data structures for quickly locating related f... |

71 | Context-based vision: Recognizing objects using information from both 2-D and 3-D imagery - Strat, Fischler - 1991 |

65 |
Geometric Invariants and Object Recognition
- Weiss
- 1993
(Show Context)
Citation Context ...ing and pose are confounded. Recently, considerable effort has been directed at identifying and employing primitives that are completely invariant with respect to viewpoint (Mundy and Zisserman 1992; =-=Weiss 1993-=-). These are based on properties of geometric structures, called invariants or geometric invariants, that remain constant over an entire class of transformations. Each is defined in terms of a particu... |

62 | Generating and generalizing models of visual objects - Connell, Brady - 1987 |

61 |
Achieving generalized object recognition through reasoning about association of function to structure
- Stark, Bowyer
- 1991
(Show Context)
Citation Context ...ring some other types of knowledge that may be used to recognize an object, such as knowledge of the context in which it may be found (Strat and Fischler 1991) or the function for which it may serve (=-=Stark and Bowyer 1991-=-). . Approaches differ according to what restrictions they place on the form of the objects recognized. The objects may be two-dimensional---like symbols printed on traffic signs---or fully three-dime... |

53 | ªEntropy and Distance of Random Graphs with Application to Structural Pattern Recognition - Wong, You - 1985 |

50 |
Criteria for Representations of Shape
- Brady
- 1983
(Show Context)
Citation Context ...r proposals for shape representations with a discussion of the criteria that ought to be satisfied by such a representation. These may be found, for example, in Marr and Nishihara 1978; Binford 1982; =-=Brady 1983-=-; Woodham 1987; Haralick, Mackworth, and Tanimoto 1988; and Mokhtarian and Mackworth 1992. The following have been mentioned frequently in some manner: 4 scope and sensitivity The representation must ... |

49 |
Survey of Model-Based Image Analysis Systems
- Binford
- 1982
(Show Context)
Citation Context ... prefaced their proposals for shape representations with a discussion of the criteria that ought to be satisfied by such a representation. These may be found, for example, in Marr and Nishihara 1978; =-=Binford 1982-=-; Brady 1983; Woodham 1987; Haralick, Mackworth, and Tanimoto 1988; and Mokhtarian and Mackworth 1992. The following have been mentioned frequently in some manner: 4 scope and sensitivity The represen... |

47 | A robot vision system for recognizing 3-D objects in low-order polynomial time
- Chen, Kak
- 1989
(Show Context)
Citation Context ...gments or patches as shape primitives, including HYPER , which uses polygonal approximations of 2D curves (Ayache and Faugeras 1986), and 3D-POLY, which uses quadric approx - imations of 3D surfaces (=-=Chen and Kak 1989-=-). b) Parts With a limited set of parts one can construct a large variety of objects, especially if each part can be customized somewhat by choosing values for free parameters. Generalized cylinders, ... |

46 |
Special Purpose Automatic Programming for 3D Model-based Vision
- Goad
- 1983
(Show Context)
Citation Context ...estimated by analyzing a model of the object. In performing this analysis some systems have simply estimated a feature's detectability as the portion of the viewsphere over which it is visible (e.g., =-=Goad 1983-=-; Kuno, Okamoto, and Okada 1991), while others have argued that much more complete models of lighting, sensors, and surfaces are needed to obtain useful estimates (e.g., Camps, Shapiro, and Haralick 1... |

45 | View variation of point-set and line-segment features - Burns, Weiss, et al. - 1993 |

45 | Finding Geometric and Relational Structures in an Image - Horaud, Skordas - 1990 |

37 |
Large hierarchical object recognition using libraries of parameterized model sub-parts
- Ettinger
- 1988
(Show Context)
Citation Context ...al finer primitives at the next (Marr and Nishihara 1978; Brooks 1981). More commonly, levels of the hierarchy represent degrees of grouping and primitives occur only at the hierarchy's lowest level (=-=Ettinger 1987-=-; Connell and Brady 1987). A second method of organizing shape primitives, not necessarily incompatible with the first, is to arrange them according to adjacency relations so that each primitive is re... |

36 |
Finding Convex Edge Groupings in an Image
- Huttenlocher, Wayner
- 1992
(Show Context)
Citation Context ...ructures for quickly locating related features (Sarkar and Boyer 1990; Saund 1990; Stein and Medioni 1992a), parallelizable algorithms (Mahoney 1987; McCafferty 1990), and other process improvements (=-=Huttenlocher and Wayner 1992-=-). Sarkar and Boyer (1993) have represented the grouping process using a Bayes network in order to allow a control strategy more flexible than the usual bottom-up process. With their approach a groupi... |

33 | Bringing the Grandmother back into the picture: A memory-based view of object recognition
- Edelman, Poggio
- 1992
(Show Context)
Citation Context ...ng has been developed using networks of neuron-like units that compute functions called generalized radial basis functions or hyper-basis functions (Poggio and Edelman 1990; Brunelli and Poggio 1991; =-=Edelman and Poggio 1990-=-). A modeled object is represented by a network in which individual units represent distinct prototypical or characteristic views of the object. The input to the network is a vector of selected image ... |

33 |
Three-dimesional recognition of solid objects from a two- dimensional image
- Huttenlocher
- 1988
(Show Context)
Citation Context ...ht for each projected feature. Because there may be many combinations of anchor feature matches, this method relies heavily on having efficient techniques for computing and verifying transformations (=-=Huttenlocher 1988-=-). The alignment method estimates a viewpoint transformation once for each set of anchor feature matches, and an error in localizing an anchor feature in the image yields an error in the transformatio... |

32 | Recognition by Prototypes - Basri - 1993 |

31 |
A framework for representing knowledge," in The psychology of computer vision
- Minsky
- 1975
(Show Context)
Citation Context ...ct views to be recorded. Thus this representation requires considerably more space than an object-centred one. Space requirements can be reduced somewhat by allowing views to share common structures (=-=Minsky 1975-=-; Burns and Riseman 1992) and by merging similar views after discarding features too fine to be reliably discerned (Petitjean, Ponce, and Kriegman 1992). Another consequence is that there are, in effe... |

31 |
Model matching in robot vision by subgraph isomorphism
- Wong
- 1992
(Show Context)
Citation Context ... complete interpretation. Correspondence space search has often been cast as a problem of graph matching (e.g., Ben-Arie and Meiri 1987; Bergevin and Levine 1993; Fan 1990; Shapiro and Haralick 1981; =-=Wong 1992-=-; Yang, Snyder, and Bilbro 1989; Zhang, Sullivan, and Baker 1992). In this framework, the task is to find a common subgraph isomorphism between two attributed graphs: one representing the image and th... |

30 | Statistical Object Recognition - Wells - 1993 |

29 |
The probabilistic peaking effect of viewed angles and distances with application to 3-D object recognition
- Ben-Arie
- 1990
(Show Context)
Citation Context ...seen from various viewpoints (Lowe 1985). Even 8 angles between general pairs of lines (not necessarily parallel ones), and ratios of line lengths, may be relatively stable with respect to viewpoint (=-=Ben-Arie 1990-=-; Burns, Weiss, and Riseman 1993). Another technique for improving the space/accuracy trade-off of a viewer-centred representation is to interpolate among views. Ullman and Basri (1991) have shown tha... |

29 |
and 0.D.Faugeras "Shape Matching of Two-Dimensional Objects
- Bhanu
- 1984
(Show Context)
Citation Context ...tes the search as soon as a solution meeting some minimum requirement has been found. Relaxation labeling has also been used as a way to shorten the search while accepting a sub-optimal result (e.g., =-=Bhanu and Faugeras 1984-=-; Bray 1990; Kitchen 1980). 3.3.2 Transformation space search The generalized Hough transform is an example of a method that searches transformation space. An array of bins, indexed by parameters of o... |

29 |
CAGD-based computer vision
- Hansen, Henderson
- 1989
(Show Context)
Citation Context ... and Kak 1989; Gottschalk, Turney, and Mudge 1989; Lamdan, Schwartz, and Wolfson 1990). Some also assess negative evidence, such as image edges that cross projected model edges at large angles (e.g., =-=Hansen and Henderson 1989-=-; Huttenlocher and Ullman 1990). Match solutions are verified by testing these measures against empirically determined thresholds, and then ranked according to the measures to select the best, mutuall... |

29 | Limitations of non model-based recognition schemes
- Moses, Ullman
- 1992
(Show Context)
Citation Context ... derivatives (e.g., Weiss 1993). It has been shown that there is no invariant for 2D projections of a finite, unconstrained set of 3D points (Clemens and Jacobs 1991a; Burns, Weiss, and Riseman 1993; =-=Moses and Ullman 1991-=-). Consequently attention has focused on invariants for suitably constrained structures, particularly coplanar sets of points, lines, and curves. Four examples will illustrate the nature of this resea... |

28 | Polynomial-time object recognition in the presence of clutter, occlusion, and uncertainty - Cass - 1992 |

27 |
Perspective projection aspect graphs of solids of revolution: an implementation
- Eggert, Bowyer
- 1991
(Show Context)
Citation Context ... analytically, the view sphere is subdivided into regions by identifying the boundaries where the object's self-occlusions begin and end, and one characteristic view is chosen from each region (e.g., =-=Eggert and Bowyer 1993-=-; Petitjean, Ponce, and Kriegman 1992). Algorithms exist for performing such analysis on a class of shapes that has been recently extended to include solids of revolution (Eggert and Bowyer 1993) and ... |

25 |
Describing and Recognizing 3-D Objects Using Surface Properties
- Fan
- 1990
(Show Context)
Citation Context ...ver the best score yet obtained for a complete interpretation. Correspondence space search has often been cast as a problem of graph matching (e.g., Ben-Arie and Meiri 1987; Bergevin and Levine 1993; =-=Fan 1990-=-; Shapiro and Haralick 1981; Wong 1992; Yang, Snyder, and Bilbro 1989; Zhang, Sullivan, and Baker 1992). In this framework, the task is to find a common subgraph isomorphism between two attributed gra... |