## Surface Light Fields for 3D Photography (2000)

### Cached

### Download Links

- [www1.cs.columbia.edu]
- [www.cs.berkeley.edu]
- [www.cs.berkeley.edu]
- [www.cs.berkeley.edu]
- [www.cs.columbia.edu]
- [www.cs.virginia.edu]
- [www.eecs.berkeley.edu]
- [www.eecs.berkeley.edu]
- [www.ics.uci.edu]
- [www.eecs.berkeley.edu]
- [www.cs.washington.edu]
- [www.wisdom.weizmann.ac.il]
- [grail.cs.washington.edu]
- [www.cs.washington.edu]
- [www.stat.washington.edu]
- [grail.cs.washington.edu]
- [classes.soe.ucsc.edu]
- [www.stat.washington.edu]
- [plenoptic.com]
- [www.stat.washington.edu]
- DBLP

### Other Repositories/Bibliography

Citations: | 159 - 1 self |

### BibTeX

@MISC{Wood00surfacelight,

author = {Daniel N. Wood and Daniel I. Azuma and Ken Aldinger and Brian Curless and Tom Duchamp and David H. Salesin and Werner Stuetzle},

title = {Surface Light Fields for 3D Photography},

year = {2000}

}

### Years of Citing Articles

### OpenURL

### Abstract

A surface light field is a function that assigns a color to each ray originating on a surface. Surface light fields are well suited to constructing virtual images of shiny objects under complex lighting conditions. This paper presents a framework for construction, compression, interactive rendering, and rudimentary editing of surface light fields of real objects. Generalizations of vector quantization and principal component analysis are used to construct a compressed representation of an object's surface light field from photographs and range scans. A new rendering algorithm achieves interactive rendering of images from the compressed representation, incorporating view-dependent geometric level-of-detail control. The surface light field representation can also be directly edited to yield plausible surface light fields for small changes in surface geometry and reflectance properties.

### Citations

2325 |
A Method for Registration of 3-D Shapes
- Besl, McKay
- 1992
(Show Context)
Citation Context ..., and the elephant from 49. Reconstructing the geometry. The scans were registered using a small number of hand-selected point correspondences to initialize a global iterated closest-points algorithm =-=[2, 10]-=-. The registered scans were merged into a single triangle mesh using the volumetric method described by Curless and Levoy [6]. The final meshes representing the surfaces of the fish and elephant conta... |

1719 |
Vector quantization and signal compression
- Gersho, Gray
- 1992
(Show Context)
Citation Context ...ecular behavior of the surface is simpler than the diffuse (e.g., an object with diffuse texture and a glossy coat.) Median removal before compression is analogous to mean-removed vector quantization =-=[11]. The seco-=-nd transformation, reflection, is a reparametrization of the lumispheres. Let n be the unit surface normal at a surface. Then for a direction � ∈ S 2 ,let�� be the reflection of � about the ... |

1144 | Progressive Meshes
- Hoppe
- 1996
(Show Context)
Citation Context ...ubtraction operations, illustrated in Figure 10(a), according to view-dependent criteria. To reduce the appearance of “popping,” we spread the visual effect of each operation over time by geomorph=-=ing [14]. In-=- a second quick pass over the mesh, we add temporary edges to eliminate cracks caused by “T-vertices,” as shown in Figure 10(b). (b)sOur criteria for wavelet addition and subtraction are the same ... |

1137 | Light field rendering
- Levoy, Hanrahan
- 1996
(Show Context)
Citation Context ...e representation, and the compression techniques they employ. Our own approach leverages high-resolution geometry to improve image quality while affording a compact representation. Levoy and Hanrahan =-=[17]-=- acquire many hundreds of images, which are resampled to lie on a regular grid in a two-plane parameterization. New images are computed by interpolation between ray samples, using essentially no geome... |

931 | Modeling and rendering architecture from photographs: A hybrid geometry- and image-based approach
- Debevec, Taylor, et al.
- 1996
(Show Context)
Citation Context ...on that can be rendered in real time. For comparable data sizes, our representation yields sharper images and greater compression ratios than two-plane representations. View-dependent texture mapping =-=[7, 8, 26]-=- is a kind of light field that does not require resampling the input images. This approach uses geometric information to re-project each input image into the desired camera viewpoint. The re-projected... |

890 | The lumigraph
- Gortler, Grzeszczuk, et al.
- 1996
(Show Context)
Citation Context ... New images are computed by interpolation between ray samples, using essentially no geometric data. They apply vector quantization to obtain compressed representations of light fields. Gortler et al. =-=[12]-=- present a similar two-plane parameterization that they call a lumigraph, in which they interpolate image samples via a hierarchical push-pull algorithm. They use approximate surface geometry derived ... |

816 | A volumetric method for building complex models from range images
- Curless, Levoy
- 1996
(Show Context)
Citation Context ...rrespondences to initialize a global iterated closest-points algorithm [2, 10]. The registered scans were merged into a single triangle mesh using the volumetric method described by Curless and Levoy =-=[6]-=-. The final meshes representing the surfaces of the fish and elephant contain 129,664 triangles and 311,376 triangles, respectively. Acquiring the photographs. We used a camera attached to a spherical... |

546 | A signal processing approach to fair surface design
- TAUBIN
- 1995
(Show Context)
Citation Context ...≡ Ns 4π � k |∆PLF(�k)| 2 . (7) The sum ranges over the vertices of the s-times-subdivided octahedron (with Ns vertices, each corresponding to a direction �k), 4 and ∆PL denotes the umbrel=-=la Laplacian [29]. Th-=-e regularized error function is then Eλ(F, Li) ≡ Edist(F, Li)+λEthin(F). (8) We use conjugate gradients to find the lumisphere F that minimizes Equation (8). Figures 3(b) and 3(c) show the faired ... |

532 | Multiresolution Analysis of Arbitrary Meshes
- ECK, DEROSE, et al.
- 1995
(Show Context)
Citation Context ...very ray originating from a surface. Our algorithm for constructing images from a surface light field relies on a good parameterization of an object’s surface mesh M. The methods of either Eck et al=-=. [9] or L-=-ee et al. [16] yield a parameterization ϕ : K0 → M ⊂ IR 3 , (1) whose domain K0 is a triangular mesh with a small number of faces, called a base mesh. We use a variant of the algorithm of Lee et ... |

403 | View-dependent Refinement of Progressive Meshes
- Hoppe
- 1997
(Show Context)
Citation Context ...se two extremes: viewdependent refinement of the subdivision-connectivity surface mesh. Most current methods for real-time, view-dependent simplification of geometry, such as those presented by Hoppe =-=[15]-=- and Xia and Varshney [32], employ progressive mesh representations and adapt the level of detail using edge collapses and vertex splits. For a texture-mapped surface, however, these operations can ca... |

360 |
An Efficient and Accurate Camera Calibration Technique for 3D Machine Vision
- Tsai
- 1986
(Show Context)
Citation Context ...ook 638 photographs of the fish and 388 photographs of the elephant, together with photographs of a calibration pattern, which we used to determine the intrinsic camera parameters using Tsai’s metho=-=d [30]-=-. During acquisition, the camera and gantry arm occasionally cast shadows onto the object. Because we wanted to capture the object under fixed lighting conditions, we manually removed photographs take... |

343 | Multiresolution Analysis for Surfaces of Arbitrary Topological Type
- LOUNSBERY, DEROSE, et al.
- 1997
(Show Context)
Citation Context ...ly near the boundaries of parameter domains, placing significant constraints on the simplification [5]. Therefore, we restrict the mesh used for rendering to have four-to-one subdivision connectivity =-=[18], and-=- refine the mesh by adding and removing lazy wavelets [4]. This allows us to modify the geometric detail almost independently of the parameterization. We approximate the map ϕ : K0 → M ⊂ IR 3 by ... |

268 | View-Dependent Simplification of Arbitrary Polygonal Environments
- Luebke, Erikson
- 1997
(Show Context)
Citation Context ...t adding significantly to the cost of projecting the error volume. Because coarse silhouettes tend to be more noticeable than interior distortion, we use a smaller error tolerance near the silhouette =-=[19]. Finall-=-y, to reduce the number of wavelet addition and subtraction operations that must be considered, we enforce one additional property: A hat function at level ℓ ≤ r, centered at an edge of Kℓ−1, ... |

238 | Efficient viewdependent image-based rendering with projective texturemapping
- Debevec, Yu, et al.
- 1998
(Show Context)
Citation Context ...on that can be rendered in real time. For comparable data sizes, our representation yields sharper images and greater compression ratios than two-plane representations. View-dependent texture mapping =-=[7, 8, 26]-=- is a kind of light field that does not require resampling the input images. This approach uses geometric information to re-project each input image into the desired camera viewpoint. The re-projected... |

232 | Multiresolution adaptive parameterization of surfaces
- LEE, SWELDENS, et al.
- 1998
(Show Context)
Citation Context ...ng from a surface. Our algorithm for constructing images from a surface light field relies on a good parameterization of an object’s surface mesh M. The methods of either Eck et al. [9] or Lee et al=-=. [16] yiel-=-d a parameterization ϕ : K0 → M ⊂ IR 3 , (1) whose domain K0 is a triangular mesh with a small number of faces, called a base mesh. We use a variant of the algorithm of Lee et al. to parameterize... |

214 | Inverse global illumination: Recovering reflectance models of real scenes from photographs
- Yu, Debevec, et al.
- 1999
(Show Context)
Citation Context ...nverse rendering is an alternative to generating a surface light field. The goal of these techniques is to estimate the surface BRDF from images and geometric data. Previous work on inverse rendering =-=[28, 33]-=- has assumed that the BRDF is piecewise linear with respect to a coarse triangulation of the surface. Our techniques require no such assumptions, and, of course, inverse rendering does not solve the r... |

195 | Object shape and reflectance modeling from observation
- Sato, Wheeler, et al.
- 1997
(Show Context)
Citation Context ...nverse rendering is an alternative to generating a surface light field. The goal of these techniques is to estimate the surface BRDF from images and geometric data. Previous work on inverse rendering =-=[28, 33]-=- has assumed that the BRDF is piecewise linear with respect to a coarse triangulation of the surface. Our techniques require no such assumptions, and, of course, inverse rendering does not solve the r... |

166 | and Amitabh Varshney. Dynamic view-dependent simplification for polygonal models
- Xia
- 1996
(Show Context)
Citation Context ...dent refinement of the subdivision-connectivity surface mesh. Most current methods for real-time, view-dependent simplification of geometry, such as those presented by Hoppe [15] and Xia and Varshney =-=[32]-=-, employ progressive mesh representations and adapt the level of detail using edge collapses and vertex splits. For a texture-mapped surface, however, these operations can cause considerable parametri... |

142 | Appearance-preserving simplification
- Cohen, Olano, et al.
- 1998
(Show Context)
Citation Context ...ture-mapped surface, however, these operations can cause considerable parametric distortion, especially near the boundaries of parameter domains, placing significant constraints on the simplification =-=[5]-=-. Therefore, we restrict the mesh used for rendering to have four-to-one subdivision connectivity [18], and refine the mesh by adding and removing lazy wavelets [4]. This allows us to modify the geome... |

113 | View-based rendering: Visualizing real objects from scanned range and color data
- PULLI, COHEN, et al.
- 1997
(Show Context)
Citation Context ...on that can be rendered in real time. For comparable data sizes, our representation yields sharper images and greater compression ratios than two-plane representations. View-dependent texture mapping =-=[7, 8, 26]-=- is a kind of light field that does not require resampling the input images. This approach uses geometric information to re-project each input image into the desired camera viewpoint. The re-projected... |

85 |
Lazy decompression of surface light fields for precomputed global illumination
- Miller, Rubin, et al.
- 1998
(Show Context)
Citation Context ...problems still need to be solved. First, we need a good representation for those 3D datasets. The framework described in this paper is based on the surface light field, a term coined by Miller et al. =-=[22]-=-. The surface light field is a function that assigns an RGB value to every ray leaving every point on a surface. When constructed from observations made of an object, a surface light field encodes suf... |

83 | Reflection Space Image Based Rendering
- Cabral, Olano, et al.
- 1999
(Show Context)
Citation Context ...bal illumination algorithm is required to produce photorealistic results. Recent work has extended interactive rendering techniques to a wider range of lighting models and environments. Cabral et al. =-=[3]-=- describe a technique for using radiance environment maps to render objects under arbitrary lighting conditions and with any isotropic BRDF. Heidrich et al. [13] use texture mapping hardware for the s... |

81 |
Hardware-accellerated Shading and Lighting
- Realistic
- 1999
(Show Context)
Citation Context ... models and environments. Cabral et al. [3] describe a technique for using radiance environment maps to render objects under arbitrary lighting conditions and with any isotropic BRDF. Heidrich et al. =-=[13]-=- use texture mapping hardware for the same purpose but allow a different class of BRDFs. However, these two methods do not handle global effects like shadows or interreflection. 1.2 Overview We have d... |

80 | Stuetzle "interactive multiresolution Surface Viewing
- Certain, Popovic, et al.
(Show Context)
Citation Context ...t constraints on the simplification [5]. Therefore, we restrict the mesh used for rendering to have four-to-one subdivision connectivity [18], and refine the mesh by adding and removing lazy wavelets =-=[4]. This a-=-llows us to modify the geometric detail almost independently of the parameterization. We approximate the map ϕ : K0 → M ⊂ IR 3 by a piecewiselinear map ϕr : Kr → IR 3 on the simplicial complex... |

73 | Eigen-texture method: Appearance compression and synthesis based on a 3d model
- NISHINO, SATO, et al.
- 2001
(Show Context)
Citation Context ...ealm that do not arise with synthetic data: most importantly, neither a surface parameterization nor the radiance along arbitrary rays are known aprioriand must instead be constructed. Nishino et al. =-=[23, 24]-=- generate surface light fields of real objects, though their images are relatively dense in only one rotational direction. Geometric information is represented by a coarse triangular mesh. They constr... |

66 | A new change of variables for efficient brdf representation. http://www.cs.princeton.edu/~smr/papers/brdf_change_of_variables
- Rusinkiewicz
- 1998
(Show Context)
Citation Context ... normal and then convolves with a “direction-invariant” filter (i.e., a space-invariant filter, where space is restricted to the surface of the sphere of directions S 2 ). As observed by Rusinkiew=-=icz [27], ma-=-ny BRDFs are approximately reflective. If these two assumptions hold, the reflected lumispheres 5 �Li and �Lj will be the same even if the normals ni and nj are different. For an example, consider... |

46 |
Registration of multiple range views for automatic 3-D modeling building
- Gagnon, Soucy, et al.
- 1994
(Show Context)
Citation Context ..., and the elephant from 49. Reconstructing the geometry. The scans were registered using a small number of hand-selected point correspondences to initialize a global iterated closest-points algorithm =-=[2, 10]-=-. The registered scans were merged into a single triangle mesh using the volumetric method described by Curless and Levoy [6]. The final meshes representing the surfaces of the fish and elephant conta... |

39 | Fitting virtual lights for non-diffuse walkthroughs
- WALTER, ALPPAY, et al.
- 1997
(Show Context)
Citation Context ... techniques to sets of texture maps. Their technique achieves compression rates for surface light fields that are comparable to those of Levoy and Hanrahan’s vector quantization method. Walter et al=-=. [31]-=- also use surface light fields to approximate solutions to global illumination problems. Their representation involves basis functions derived from hardware lighting models, which provides very fast r... |

22 | Appearance compression and synthesis based on 3d model for mixed reality
- NISHINO, SATO, et al.
- 1999
(Show Context)
Citation Context ...ealm that do not arise with synthetic data: most importantly, neither a surface parameterization nor the radiance along arbitrary rays are known aprioriand must instead be constructed. Nishino et al. =-=[23, 24]-=- generate surface light fields of real objects, though their images are relatively dense in only one rotational direction. Geometric information is represented by a coarse triangular mesh. They constr... |

21 | Hierarchical coding of light fields with disparity maps
- Magnor, Girod
- 1999
(Show Context)
Citation Context ...rection that substantially reduces ghosting and blurring artifacts. In both these methods, the representation restricts the viewpoint to lie outside of the convex hull of the object. Magnor and Girod =-=[20, 21]-=- develop an MPEG-like scheme for compressing two-plane light fields that produces better compression ratios than those obtained by Levoy and Hanrahan. Our approach depends on both highresolution geome... |

15 | Adaptive block-based light field coding
- Magnor, Girod
- 1999
(Show Context)
Citation Context ...rection that substantially reduces ghosting and blurring artifacts. In both these methods, the representation restricts the viewpoint to lie outside of the convex hull of the object. Magnor and Girod =-=[20, 21]-=- develop an MPEG-like scheme for compressing two-plane light fields that produces better compression ratios than those obtained by Levoy and Hanrahan. Our approach depends on both highresolution geome... |

2 |
Interactive rendering of surface light fields
- Azuma
- 2000
(Show Context)
Citation Context ...d Levoy and Hanrahan. We have observed that lowering this resolution results in artifacts such as erroneous interpolation among rays that strike quite different surface points due to occlusion. Azuma =-=[1]-=- discusses this effect and other difficulties inherent in reduction of the near-plane resolution. 5 Rendering In this section we present an interactive surface light rendering algorithm. Our implement... |