## Feature extraction, image segmentation and surface fitting: the development of a 3D scene reconstruction system (1998)

Citations: | 3 - 0 self |

### BibTeX

@TECHREPORT{Lester98featureextraction,,

author = {Eric D. Lester},

title = {Feature extraction, image segmentation and surface fitting: the development of a 3D scene reconstruction system },

institution = {},

year = {1998}

}

### OpenURL

### Abstract

### Citations

3109 | A computational approach to edge detection
- Canny
- 1986
(Show Context)
Citation Context ...re, di erent approaches are used for each data type. There are twotypes of features we detect in images. They are crease edges and step edges. The classical works in edge detection are those of Canny =-=[3]-=- and Marr & Hildreth [4]. However, there are a variety of other methods used for feature extraction in intensity and amplitude images. Boyer and Sarker [5] give an assessment of the state of the art i... |

2249 |
A mathematical theory of evidence
- Shafer
- 1976
(Show Context)
Citation Context ...hrough 2.46. Fuzzy Intersection a\b: Dubois & Prade [77] a b max(a; b; ) 32 (2:42)sDombi [77] Fuzzy Union a[b: Dubois & Prade [77] Dombi [77] where2 (0; 1) and2 (0;1). Bernoulli's rule of combination =-=[78]-=- 1 1+[( 1 a , 1) +(1 b , 1) ] 1 (2:43) a + b,a b,min(a; b; 1, ) max(1, a; 1, b; ) (2:44) 1 1+[( 1 a , 1), +( 1 b , 1), ] ,1 (2:45) 1,(1,a) (1, b) (2:46) Given we are fusing data that may not be redund... |

1287 | Scale-space and edge detection using anisotropic diffusion
- Perona, Malik
- 1990
(Show Context)
Citation Context ...mensional function of x and y. Therefore, for a single image, the general form of the smoothing function is @f(x; y; t) @t =r g(x; y; t)rf(x; y; t): (2:9) Where g(x,y,t) = is the conductance function =-=[76]-=-. We incorporate multiple images in the anisotropic di usion smoothing function. For multiple images, the smoothing function is @ ~ f(x; y; t) @t =r g(x; y; t)r ~ f(x; y; t): (2:10) In general, ~ f(x;... |

854 | Theory of edge detection
- Marr, Hildreth
- 1980
(Show Context)
Citation Context ...are used for each data type. There are twotypes of features we detect in images. They are crease edges and step edges. The classical works in edge detection are those of Canny [3] and Marr & Hildreth =-=[4]-=-. However, there are a variety of other methods used for feature extraction in intensity and amplitude images. Boyer and Sarker [5] give an assessment of the state of the art in edge detection in thes... |

796 | A volumetric method for building complex models from range images
- Curless, Levoy
- 1996
(Show Context)
Citation Context ...t to the surface of the data like a balloon. Other methods are implemented to build complex models from range images based on volumetric occupancy grids and signed distance functions and are given in =-=[56]-=- [57] [58]. A good review of 3D surface tting is given in [59]. 12s1.5 Review of Scene Reconstruction Systems We de ne scene reconstruction as inferring models of a scene from a set of noisy measureme... |

496 |
Characterization of signals from multiscale edges
- Mallat, Zhong
- 1992
(Show Context)
Citation Context ...es by applying larger lters. The goal is to obtain an appropriate scale which is smooth enough for di erentiation but does not introduce too much blurring. Other work in this area may be seen in [18] =-=[19]-=- [20]. The fourth category of range image feature extraction is di erential geometry. Differential geometry in range images is locating and classifying geometric properties of the scene by means of es... |

454 |
Zippered Polygon Meshes from Range Images
- Turk, Levoy
- 1994
(Show Context)
Citation Context ...equeira, Goncalves, and Ribeiro [64] present an approach which combines various range images of a single scene from di ering viewpoints to develop high-level descriptions of the scene. Turk and Levoy =-=[65]-=- combine multiple range images to create a zippered polygon mesh of an object in a scene so that the object may be rotated and viewed from any angle. Zhang and Wallace [66] present an approach similar... |

350 |
Dynamic 3d models with local and global deformations: deformable superquadrics
- Terzopoulos, Metaxas
- 1991
(Show Context)
Citation Context ...s of the volumetric shapes and the surface shapes are de ned by the sign minimum and maximum curvatures of the individual face. Other examples of this technique with various applications are given in =-=[37]-=- [38] [39] [40]. The third division of range image segmentation techniques is the set of hybrid approaches which combine both edge and surface properties to obtain a nal segmentation. A classic exampl... |

288 |
A review on image segmentation techniques
- Pal, Pal
- 1993
(Show Context)
Citation Context ...mage into homogeneous regions, or patches, each of which corresponds to a continuous portion of that image. There are a variety of methods used to segment both range and intensity images. Pal and Pal =-=[26]-=- give a somewhat comprehensive review of image segmentation techniques including intensity images. Our review concentrates primarily on range image segmentation. Wani and Batchelor [27] state that, in... |

267 |
Image processing handbook
- Russ
- 2002
(Show Context)
Citation Context ...ent, has relatively few free parameters, and o ers a means of combining di erent types of image features. Other background material and examples of this type of segmentation may be found in [43] [44] =-=[45]-=- [42] [14] [46]. 1.4 Review of Surface Fitting The objective of surface tting is to describe objects in a scene using mathematical models. There are a number of di erent methods for surface tting. The... |

227 | An Experimental Comparison or range image segmentation algorithms
- Hoover, Jean-Baptiste, et al.
(Show Context)
Citation Context ...an initial segmentation. Then these regions are 8smerged based on the similarity of their quadric t parameters. Four individual methods implemented using this type of general approach may be found in =-=[30]-=-. Other methods include [31] [32] [33] [34] [35]. The second category is model-based segmentation. Model-based segmentation begins with a set of three-dimensional models and segments the range image b... |

227 |
Estimation of planar curves, surfaces and nonplanar space curves defined by implicit equations, with applications to edge and range image segmentation
- Taubin
- 1991
(Show Context)
Citation Context ...se regions are 8smerged based on the similarity of their quadric t parameters. Four individual methods implemented using this type of general approach may be found in [30]. Other methods include [31] =-=[32]-=- [33] [34] [35]. The second category is model-based segmentation. Model-based segmentation begins with a set of three-dimensional models and segments the range image based on the t of the data to the ... |

198 |
Segmentation Through Variable-Order Surface Fitting
- Besl, Jain
- 1988
(Show Context)
Citation Context ...n these regions are 8smerged based on the similarity of their quadric t parameters. Four individual methods implemented using this type of general approach may be found in [30]. Other methods include =-=[31]-=- [32] [33] [34] [35]. The second category is model-based segmentation. Model-based segmentation begins with a set of three-dimensional models and segments the range image based on the t of the data to... |

183 |
Shape and nonrigid motion estimation throughphysics-basedsynthesis
- Metaxas, Terzopoulos
- 1993
(Show Context)
Citation Context ...olumetric shapes and the surface shapes are de ned by the sign minimum and maximum curvatures of the individual face. Other examples of this technique with various applications are given in [37] [38] =-=[39]-=- [40]. The third division of range image segmentation techniques is the set of hybrid approaches which combine both edge and surface properties to obtain a nal segmentation. A classic example of this ... |

151 | A level-set approach to 3D reconstruction from range data
- Whitaker
- 1988
(Show Context)
Citation Context ...ent and reconstruct a scene based on a single range image. Gokmen and Li [8] use re ned regularization where 13sthe smoothness of of the surface is regulated spatially over the entire image. Whitaker =-=[68]-=- proposes using level-set models to reconstruct a 3D scene using more than one point of view. One of the most recent advances of scene reconstruction is the development of the Artisan system by The Ro... |

104 |
Using deformable surfaces to segment 3-D images and infer differential structures
- Cohen, Cohen, et al.
- 1992
(Show Context)
Citation Context ...tric shapes and the surface shapes are de ned by the sign minimum and maximum curvatures of the individual face. Other examples of this technique with various applications are given in [37] [38] [39] =-=[40]-=-. The third division of range image segmentation techniques is the set of hybrid approaches which combine both edge and surface properties to obtain a nal segmentation. A classic example of this appro... |

103 |
On three-dimensional surface reconstruction methods
- Bolle, Vemuri
- 1991
(Show Context)
Citation Context ... implemented to build complex models from range images based on volumetric occupancy grids and signed distance functions and are given in [56] [57] [58]. A good review of 3D surface tting is given in =-=[59]-=-. 12s1.5 Review of Scene Reconstruction Systems We de ne scene reconstruction as inferring models of a scene from a set of noisy measurements (range or intensity). Scene reconstruction is used for a v... |

98 | Dynamic deformation of solid primitives with constraints
- METAXAS, TERZOPOULOS
- 1992
(Show Context)
Citation Context ...the volumetric shapes and the surface shapes are de ned by the sign minimum and maximum curvatures of the individual face. Other examples of this technique with various applications are given in [37] =-=[38]-=- [39] [40]. The third division of range image segmentation techniques is the set of hybrid approaches which combine both edge and surface properties to obtain a nal segmentation. A classic example of ... |

95 | A robust visual method for assessing the relative performance of edge detection algorithms
- Heath, Sarkar, et al.
- 1997
(Show Context)
Citation Context ... applying larger lters. The goal is to obtain an appropriate scale which is smooth enough for di erentiation but does not introduce too much blurring. Other work in this area may be seen in [18] [19] =-=[20]-=-. The fourth category of range image feature extraction is di erential geometry. Differential geometry in range images is locating and classifying geometric properties of the scene by means of estimat... |

67 |
Darboux frames, snakes, and superquadrics: Geometry from the bottom up
- Ferrie, Lagarde, et al.
- 1993
(Show Context)
Citation Context ...ge image, or viewpoint of the scene. Taubin [50] proposes an improved method of surface tting by using a better approximation of the Euclidean distance of the surface to the actual data. Ferrie et al =-=[51]-=- cover three approaches which are used in three-dimensional surface tting and explain the use of di erential geometry in volumetric modeling. Using this basic strategy, Kumar et al [52] use hyperquadr... |

67 |
Describing complicated objects by implicit polynomials
- Keren, Cooper, et al.
- 1994
(Show Context)
Citation Context ...escribe volumes in range data using a physics-based framework in the form of blending by allowing di erent geometric models combine to represent a single object in a uni ed model. Lastly, Keren et al =-=[54]-=- describe complex objects in a scene by fourth-degree polynomials and claim that they are mathematically easier to manipulate than superquadrics. Another category of three-dimensional surface tting st... |

48 | Describing surfaces
- Brady, Ponce, et al.
- 1985
(Show Context)
Citation Context ...d on these adjacent depths. This is the approach which wehave taken in our feature extraction techniques for range images. Other examples of work that rely on di erential geometry may be seen in [23] =-=[24]-=-. Of course these four categories of feature detection are not always mutually exclusive. Techniques used in each category may be combined and even used with others. For instance, Burgiss's work [18],... |

48 |
An improved algorithm for algebraic curve and surface fitting
- Taubin
- 1993
(Show Context)
Citation Context ...een the proposed surface and the actual data points. 11sThree-dimensional surface tting can be divided into two categories. The rst is based on a single range image, or viewpoint of the scene. Taubin =-=[50]-=- proposes an improved method of surface tting by using a better approximation of the Euclidean distance of the surface to the actual data. Ferrie et al [51] cover three approaches which are used in th... |

33 |
Morphological Edge Detection
- Lee, Haralick, et al.
- 1987
(Show Context)
Citation Context ... for a negative roof edge, closing renders a nonzero response. The basic morphological operators and their operations are given in [13] [14]. Other techniques based on this approach are given in [15] =-=[16]-=-. The third category is multi-scale feature extraction. Multi-scale feature extraction used neighborhood operators of various sizes for smoothing or polynomial approximation. In general, the larger th... |

28 |
The robust sequential estimator: A general approach and its application to surface organization in range data
- Boyer, Mirza, et al.
- 1994
(Show Context)
Citation Context ... Typically the solution to these types of ts are much simpler than in the three-dimensional case and lend themselves to closed-form solutions. They may be quadratics or other polynomials. Boyer et al =-=[47]-=- present such an approach based on the robust sequential estimator which is able to handle handle noise and remove outliers in the process of estimating the surface. Miller and Stewart [48] propose a ... |

28 |
Shape evolution with structural and topological changes using blending
- DeCarlo, Metaxas
- 1998
(Show Context)
Citation Context ...ting and explain the use of di erential geometry in volumetric modeling. Using this basic strategy, Kumar et al [52] use hyperquadrics to describe the range data in a given scene. DeCarlo and Metaxas =-=[53]-=- describe volumes in range data using a physics-based framework in the form of blending by allowing di erent geometric models combine to represent a single object in a uni ed model. Lastly, Keren et a... |

25 |
Edge-region-based segmentation of range images
- Wani, Batchelor
(Show Context)
Citation Context ...s. Pal and Pal [26] give a somewhat comprehensive review of image segmentation techniques including intensity images. Our review concentrates primarily on range image segmentation. Wani and Batchelor =-=[27]-=- state that, in general, three-dimensional image segmentation techniques may be categorized under two main classes: region growing methods, and edgebased methods. Bhandarkar and Siebert [28] state tha... |

24 | The role of model-based segmentation in the recovery of volumetric parts from range data
- Dickinson, Metaxas, et al.
- 1997
(Show Context)
Citation Context ...ge based on the t of the data to the given model. This approach relates closely to some of the work done in surface tting, but it appears to be useful for segmentation. An example of this is shown in =-=[36]-=- where they begin with ten modeling primitives and recover the volumetric parts from range data by using a hierarchical segmentation by mapping boundary groups to faces, faces to aspects, and aspects ... |

21 |
On recovering hyperquadrics from range data
- Kumar, Han, et al.
- 1995
(Show Context)
Citation Context .... Ferrie et al [51] cover three approaches which are used in three-dimensional surface tting and explain the use of di erential geometry in volumetric modeling. Using this basic strategy, Kumar et al =-=[52]-=- use hyperquadrics to describe the range data in a given scene. DeCarlo and Metaxas [53] describe volumes in range data using a physics-based framework in the form of blending by allowing di erent geo... |

20 |
Segmentation of 3D range images using pyramidal data structures
- Arman, Sabata, et al.
- 1993
(Show Context)
Citation Context ...of the above mentioned characteristics at adjacent pixels. The result is a range image segmented into regions that contain the same properties. An example of this type of segmentation is Sabata et al =-=[29]-=-, in which the scene is originally segmented into homogeneous regions by clustering surface normals to obtain an initial segmentation. Then these regions are 8smerged based on the similarity of their ... |

20 | A system for semi-automatic modeling of complex environment
- Johnson, Hoffman, et al.
- 1997
(Show Context)
Citation Context ... a 3D scene using more than one point of view. One of the most recent advances of scene reconstruction is the development of the Artisan system by The Robotics Institute at Carnegie Mellon University =-=[69]-=-. The Artisan system incorporates multiple range views and reconstructs the scene piece by piece, via ahuman operator, by matching CAD models to selected regions of interest, or objects. The results a... |

18 |
Fast segmentation of range imagery into planar regions
- Taylor, Savini, et al.
- 1989
(Show Context)
Citation Context ...8smerged based on the similarity of their quadric t parameters. Four individual methods implemented using this type of general approach may be found in [30]. Other methods include [31] [32] [33] [34] =-=[35]-=-. The second category is model-based segmentation. Model-based segmentation begins with a set of three-dimensional models and segments the range image based on the t of the data to the given model. Th... |

17 |
Edge detection and surface reconstruction using re ned regularization
- Gokmen, Li
- 1993
(Show Context)
Citation Context ...der stabilizer capable of smoothing heavily in areas of low contrast and performing little or no smoothing in areas of high gradient, i.e. edges. Other examples of this typeofwork may be found in [7] =-=[8]-=- [9] [10] [11]. A second category is mathematical morphology. In this approach, morphological oper5sators such as erosion, dilation, openings, and closings are used to detect and classify edge types i... |

15 |
Experimental characterization of the perceptron laser range nder
- Kweon, man, et al.
- 1991
(Show Context)
Citation Context ...f so desired. However, we are only able to obtain a range image of the scene and not a registered pair including amplitude data. The real data is obtained using the Perceptron laser range scanner [1] =-=[74]-=- that we have in house. This laser scanner has a 55 eld of view on the horizontal axis and a 45 eld of view on the vertical axis. We work with images that vary in size from 256 256 pixels to 1024 1024... |

14 |
Numerical Recipes in C: The Art of Scienti c Computing. Second Edition
- Press, Teukolsky, et al.
- 1992
(Show Context)
Citation Context ...xx 2 + Cxyxy + Cyyy 2 : (4:2) 47sThe coe cients may be represented in vector form as ~C = 2 64 Cx Cy Cxx Cyy Cxy 3 75 : (4:3) This linear least-squares t is achieved by a singular value decomposition =-=[80]-=-. We classify the surface based on the di erential properties at the middle of the patch. The derivatives of f(x; y) are fx =2CxxM[~xi]+CxyM[~yi]+Cx (4:4) fy =2CyyM[~yi]+CxyM[~xi]+Cy fxx =2Cxx fyy =2C... |

13 |
Image Relaxation: Restoration and Feature Extraction
- Snyder, Han, et al.
- 1995
(Show Context)
Citation Context ...r capable of smoothing heavily in areas of low contrast and performing little or no smoothing in areas of high gradient, i.e. edges. Other examples of this typeofwork may be found in [7] [8] [9] [10] =-=[11]-=-. A second category is mathematical morphology. In this approach, morphological oper5sators such as erosion, dilation, openings, and closings are used to detect and classify edge types in range data. ... |

11 | Segmentation of range images via data fusion and morpholocial watersheds. Pattern Recognition
- Baccar, Gee, et al.
- 1996
(Show Context)
Citation Context ...vely few free parameters, and o ers a means of combining di erent types of image features. Other background material and examples of this type of segmentation may be found in [43] [44] [45] [42] [14] =-=[46]-=-. 1.4 Review of Surface Fitting The objective of surface tting is to describe objects in a scene using mathematical models. There are a number of di erent methods for surface tting. These methods mayb... |

10 |
L.: Sensor integration for robot navigation: Combining sonar and stereo range data in a grid-based representation
- Elfes, Matthies
- 1987
(Show Context)
Citation Context ...the surface of the data like a balloon. Other methods are implemented to build complex models from range images based on volumetric occupancy grids and signed distance functions and are given in [56] =-=[57]-=- [58]. A good review of 3D surface tting is given in [59]. 12s1.5 Review of Scene Reconstruction Systems We de ne scene reconstruction as inferring models of a scene from a set of noisy measurements (... |

10 |
Physical modeling and combination of range and intensity edge data
- Zhang, Wallace
- 1993
(Show Context)
Citation Context ...f the scene. Turk and Levoy [65] combine multiple range images to create a zippered polygon mesh of an object in a scene so that the object may be rotated and viewed from any angle. Zhang and Wallace =-=[66]-=- present an approach similar to ours in that they combine range and intensity edge information to physically model a given scene. Di erent methods may be used to accomplish the task of scene reconstru... |

8 | Range image segmentation through pattern analysis of the multiscale wavelet transform
- Burgiss, Whitaker, et al.
- 1998
(Show Context)
Citation Context ...eatures by applying larger lters. The goal is to obtain an appropriate scale which is smooth enough for di erentiation but does not introduce too much blurring. Other work in this area may be seen in =-=[18]-=- [19] [20]. The fourth category of range image feature extraction is di erential geometry. Differential geometry in range images is locating and classifying geometric properties of the scene by means ... |

8 |
Description of 3-d surfaces using curvature properties
- Medioni, Nevatia
- 1984
(Show Context)
Citation Context ... based on these adjacent depths. This is the approach which wehave taken in our feature extraction techniques for range images. Other examples of work that rely on di erential geometry may be seen in =-=[23]-=- [24]. Of course these four categories of feature detection are not always mutually exclusive. Techniques used in each category may be combined and even used with others. For instance, Burgiss's work ... |

8 |
Integrating edge and surface information for range image segmentation
- Bhandarkar, Siebert
- 1992
(Show Context)
Citation Context ...Batchelor [27] state that, in general, three-dimensional image segmentation techniques may be categorized under two main classes: region growing methods, and edgebased methods. Bhandarkar and Siebert =-=[28]-=- state that most range image segmentation algorithms could be classi ed as either surface-based or edge-based. These two general classi cations are typical of what we have seen in the literature. Howe... |

7 | Adaptive Multiscale Feature Extraction from Range Data
- Parvin, Medioni
- 1989
(Show Context)
Citation Context ...he image and interferes with the feature extraction technique. Thus, by using multiple scales and combining them, a more complete analysis of the features present may be performed. Parvin and Medioni =-=[17]-=- propose an adaptive multi-scale method in which they select an appropriate scale which would be minimum in size while producing a maxima of the curvature at locations of the zero-crossing of the rst ... |

7 |
3D environment modeling using laser range sensing
- Sequeira, Goncalves, et al.
- 1995
(Show Context)
Citation Context ...60] propose using scene reconstruction for robotic inspection and manipulation of objects in an area. Jung and Gupta [61] propose using scene reconstruction for path planning purposes. Sequeira et al =-=[62]-=- uses solely laser range sensing to model, or reconstruct a 3D environment for navigation and veri cation purposes. Walker [63] presents a method that uses the geometric properties of objects in a sce... |

5 | Segmentation of range images using morphological operations: review and examples
- Gee, Abidi
- 1995
(Show Context)
Citation Context ... For a positive roof edge, opening gives a nonzero response, and for a negative roof edge, closing renders a nonzero response. The basic morphological operators and their operations are given in [13] =-=[14]-=-. Other techniques based on this approach are given in [15] [16]. The third category is multi-scale feature extraction. Multi-scale feature extraction used neighborhood operators of various sizes for ... |

4 |
Fusion of Multi-dimensional Data Using Regularization
- Abidi
- 1992
(Show Context)
Citation Context ...ilizer capable of smoothing heavily in areas of low contrast and performing little or no smoothing in areas of high gradient, i.e. edges. Other examples of this typeofwork may be found in [7] [8] [9] =-=[10]-=- [11]. A second category is mathematical morphology. In this approach, morphological oper5sators such as erosion, dilation, openings, and closings are used to detect and classify edge types in range d... |

4 | Autonomous robotics inspection and manipulation using multisensor feedback
- Abidi, Eason, et al.
- 1991
(Show Context)
Citation Context ...ruction Systems We de ne scene reconstruction as inferring models of a scene from a set of noisy measurements (range or intensity). Scene reconstruction is used for a variety of purposes. Abidi et al =-=[60]-=- propose using scene reconstruction for robotic inspection and manipulation of objects in an area. Jung and Gupta [61] propose using scene reconstruction for path planning purposes. Sequeira et al [62... |

4 | High-Level Surface Descriptions from Composite Range Images
- Sequeira, Goncalves, et al.
- 1995
(Show Context)
Citation Context ...face tting, many systems utilize only one source of input data to reconstruct the scene. Other systems utilize multi-sensor or multi-view feedback for reconstruction. Sequeira, Goncalves, and Ribeiro =-=[64]-=- present an approach which combines various range images of a single scene from di ering viewpoints to develop high-level descriptions of the scene. Turk and Levoy [65] combine multiple range images t... |

3 |
calibration and use of the perceptron laser range nder in a controlled enviroment
- Pito, \Characterization
- 1995
(Show Context)
Citation Context ...r range scanner representing the distance from the source of the laser to the nearest object along the line of sight. Amplitude data is the magnitude of the signal received from a laser range scanner =-=[1]-=-. Intensity data is the re ectance of an object in a scene received by a passive sensor such as a CCD array camera or standard video camera. Figure 1.1 shows these three types of images. Note that vis... |

3 |
Surface characterization using a Gaussian weighted least squares technique towards segmentation of range images
- Baccar
- 1994
(Show Context)
Citation Context ...ture map created by the fusion of di ering edge types and changes in the 9ssurface normal, then performing a watershed algorithm on the nal fused input feature map. Such an example is given by Baccar =-=[42]-=- in which a Gaussian weighted least squares technique is proposed to extract step and crease edges, then combine them using a special case of Dempster-Shafer fusion, Bernoulli's rule of combination, a... |