## Fast texture synthesis using tree-structured vector quantization (2000)

### Cached

### Download Links

- [www.cs.stevens.edu]
- [graphics.stanford.edu]
- [www-graphics.stanford.edu]
- [graphics.stanford.edu]
- [www.cs.brown.edu]
- [www-hci.stanford.edu]
- [www-graphics.stanford.edu]
- [graphics.stanford.edu]
- [cgm.cs.ntust.edu.tw]
- [graphics.stanford.edu]
- [www-graphics.stanford.edu]
- [www.visgraf.impa.br]
- DBLP

### Other Repositories/Bibliography

Citations: | 450 - 12 self |

### BibTeX

@INPROCEEDINGS{Wei00fasttexture,

author = {Li-yi Wei and Marc Levoy},

title = {Fast texture synthesis using tree-structured vector quantization},

booktitle = {},

year = {2000},

pages = {479--488}

}

### Years of Citing Articles

### OpenURL

### Abstract

Figure 1: Our texture generation process takes an example texture patch (left) and a random noise (middle) as input, and modifies this random noise to make it look like the given example texture. The synthesized texture (right) can be of arbitrary size, and is perceived as very similar to the given example. Using our algorithm, textures can be generated within seconds, and the synthesized results are always tileable. Texture synthesis is important for many applications in computer graphics, vision, and image processing. However, it remains difficult to design an algorithm that is both efficient and capable of generating high quality results. In this paper, we present an efficient algorithm for realistic texture synthesis. The algorithm is easy to use and requires only a sample texture as input. It generates textures with perceived quality equal to or better than those produced by previous techniques, but runs two orders of magnitude faster. This permits us to apply texture synthesis to problems where it has traditionally been considered impractical. In particular, we have applied it to constrained synthesis for image editing and temporal texture generation. Our algorithm is derived from Markov Random Field texture models and generates textures through a deterministic searching process. We accelerate this synthesis process using tree-structured vector quantization.

### Citations

1774 | Vector quantization and signal compression - Gersho, Gray - 1992 |

817 | Texture synthesis by non-parametric sampling
- Efros, Leung
- 1999
(Show Context)
Citation Context ...es. Markov Random Field and Gibbs Sampling: Many algorithms model textures by Markov Random Fields (or in a different mathematical form, Gibbs Sampling), and generate textures by probability sampling =-=[6, 28, 20, 18]-=-. Since Markov Random Fields have been proven to be a good approximation for a broad range of textures, these algorithms are general and some of them produce good results. A drawback of Markov Random ... |

676 | Textures: a Photographic Album for Artists and Designers - Brodatz - 1965 |

427 | The digital michelangelo project: 3D scanning of large statues
- Levoy
- 2000
(Show Context)
Citation Context ...resulting displacement maps should be compressable/decompressable as 2D textures using our technique. Taking this idea further, missing geometric details, a common problem in many scanning situations =-=[14]-=-, could be filled in using our constrained texture synthesis technique. Direct synthesis over meshes: Mapping textures onto irregular 3D meshes by projection often causes distortions [21]. These disto... |

400 | Pyramid-based texture analysis/synthesis
- Heeger, Bergen
- 1995
(Show Context)
Citation Context ...small texture patches can take hours or days to generate. Feature Matching: Some algorithms model textures as a set of features, and generate new images by matching the features in an example texture =-=[9, 4, 22]-=-. These algorithms are usually more efficient than Markov Random Field algorithms. Heeger and Bergen [9] model textures by matching marginal histograms of image pyramids. Their technique succeeds on h... |

287 | A multiresolution spline with application to image mosaics
- Burt, Adelson
- 1983
(Show Context)
Citation Context ...for textures containing large scale structures we have to use large neighborhoods, and large neighborhoods demand more computation. This problem can be solved by using a multiresolution image pyramid =-=[3]-=-; computation is saved because we can represent large scale structures more compactly by a few pixels in a certain lower resolution pyramid level. The multiresolution synthesis algorithm proceeds as f... |

243 | Multiresolution sampling procedure for analysis and synthesis of texture image
- Bonet
- 1997
(Show Context)
Citation Context ...small texture patches can take hours or days to generate. Feature Matching: Some algorithms model textures as a set of features, and generate new images by matching the features in an example texture =-=[9, 4, 22]-=-. These algorithms are usually more efficient than Markov Random Field algorithms. Heeger and Bergen [9] model textures by matching marginal histograms of image pyramids. Their technique succeeds on h... |

224 | Fitting Smooth Surfaces to Dense Polygon Meshes. Computer Graphics
- Krishnamurthy, Levoy
(Show Context)
Citation Context ...orhood vectors to build the (full) codebooks. models expensive to store, transmit or manipulate. These geometric details can be represented as displacement maps over a smoother surface representation =-=[13]-=-. The resulting displacement maps should be compressable/decompressable as 2D textures using our technique. Taking this idea further, missing geometric details, a common problem in many scanning situa... |

215 |
P.: Fast Shadows and Lighting Effects using Texture Mapping
- Segal, Korobkin, et al.
- 1992
(Show Context)
Citation Context ...g situations [14], could be filled in using our constrained texture synthesis technique. Direct synthesis over meshes: Mapping textures onto irregular 3D meshes by projection often causes distortions =-=[21]-=-. These distortions can sometimes be fixed by establishing suitable parameterization of the mesh, but a more direct approach would be to synthesize the texture directly over the mesh. In principle, th... |

134 | A simple algorithm for nearest neighbor search in high dimensions
- Nene, Nayar
- 1997
(Show Context)
Citation Context ...celeration is possible. This is achieved by considering neighborhoods N(p) as points in a multiple dimensional space, and casting the neighborhood matching process as a nearestpoint searching problem =-=[17]-=-. The nearest-point searching problem in multiple dimensions is stated as follows: given a set S of n points and a novel query point Q in a d-dimensional space, find a point in the set such that its d... |

133 | Reaction-diffusion textures
- Witkin, Kass
- 1991
(Show Context)
Citation Context ...s possible to synthesize certain surface textures by directly simulating their physical generation processes. Biological patterns such as fur, scales, and skin can be modeled using reaction diffusion =-=[26]-=- and cellular texturing [27]. Some weathering and mineral phenomena can be faithfully reproduced by detailed simulations [5]. These techniques can produce textures directly on 3D meshes so the texture... |

132 | Depicting fire and other gaseous phenomena using diffusion processes
- Stam, Fiume
- 1995
(Show Context)
Citation Context ...s of computer graphics, a technique that can synthesize temporal textures would be useful. Most existing algorithms model temporal textures by direct simulation; examples include fluid, gas, and fire =-=[23]-=-. Direct simulations, however, are often expensive and only suitable for specific kinds of textures; therefore an algorithm that can model general motion textures would be advantageous [24]. Temporal ... |

121 | Temporal Texture Modeling
- Szummer, Picard
- 1995
(Show Context)
Citation Context ..., and fire [23]. Direct simulations, however, are often expensive and only suitable for specific kinds of textures; therefore an algorithm that can model general motion textures would be advantageous =-=[24]-=-. Temporal textures consist of 3D spatial-temporal volume of motion data. If the motion data is local and stationary both in space and time, the texture can be synthesized by a 3D extension of our alg... |

107 | Texture characterization via joint statistics of wavelet coefficient magnitudes
- Simoncelli, Portilla
(Show Context)
Citation Context ...small texture patches can take hours or days to generate. Feature Matching: Some algorithms model textures as a set of features, and generate new images by matching the features in an example texture =-=[9, 4, 22]-=-. These algorithms are usually more efficient than Markov Random Field algorithms. Heeger and Bergen [9] model textures by matching marginal histograms of image pyramids. Their technique succeeds on h... |

99 | Modeling and rendering of weathered stone
- Dorsey, Edelman, et al.
- 1999
(Show Context)
Citation Context ...erns such as fur, scales, and skin can be modeled using reaction diffusion [26] and cellular texturing [27]. Some weathering and mineral phenomena can be faithfully reproduced by detailed simulations =-=[5]-=-. These techniques can produce textures directly on 3D meshes so the texture mapping distortion problem is avoided. However, different textures are usually generated by very different physical process... |

99 |
Cellular texture basis function
- Worley
- 1996
(Show Context)
Citation Context ...tain surface textures by directly simulating their physical generation processes. Biological patterns such as fur, scales, and skin can be modeled using reaction diffusion [26] and cellular texturing =-=[27]-=-. Some weathering and mineral phenomena can be faithfully reproduced by detailed simulations [5]. These techniques can produce textures directly on 3D meshes so the texture mapping distortion problem ... |

72 | Novel cluster-based probability model for texture synthesis, classification, and compression
- Popat, Picard
- 1993
(Show Context)
Citation Context ...es. Markov Random Field and Gibbs Sampling: Many algorithms model textures by Markov Random Fields (or in a different mathematical form, Gibbs Sampling), and generate textures by probability sampling =-=[6, 28, 20, 18]-=-. Since Markov Random Fields have been proven to be a good approximation for a broad range of textures, these algorithms are general and some of them produce good results. A drawback of Markov Random ... |

53 | Rendering from compressed textures
- Beers, Agrawala, et al.
- 1996
(Show Context)
Citation Context ...eating patterns and high frequency information; therefore they are not well compressed by transform-based techniques such as JPEG. However, codebook-based compression techniques work well on textures =-=[1]-=-. This suggests that textures might be compressable by our synthesis technique. Compression would consist of building a codebook, but unlike [1], no code indices would be generated; only the codebook ... |

50 | Image replacement through texture synthesis
- Igehy, Pereira
- 1997
(Show Context)
Citation Context ...ractive algorithm that finds translationally similar regions for noise removal. Often, the flawed portion is contained within a region of texture, and can be replaced by constrained texture synthesis =-=[6, 11]-=-. Texture replacement by constrained synthesis must satisfy two requirements: the synthesized region must look like the surrounding texture, and the boundary between the new and old regions must be in... |

50 | Texture synthesis via a noncausal nonparametric multiscale Markov random field
- Pagete, Longstaff
- 1998
(Show Context)
Citation Context ...es. Markov Random Field and Gibbs Sampling: Many algorithms model textures by Markov Random Fields (or in a different mathematical form, Gibbs Sampling), and generate textures by probability sampling =-=[6, 28, 20, 18]-=-. Since Markov Random Fields have been proven to be a good approximation for a broad range of textures, these algorithms are general and some of them produce good results. A drawback of Markov Random ... |

47 | Combining Frequency and spatial domain information for fast interactive image noise removal. Computer Graphics
- Hirani, Totsuka
- 1996
(Show Context)
Citation Context ...frame, or simply an undesirable object in an image. Since the processes causing these flaws are often irreversible, an algorithm that can fix these flaws is desirable. For example, Hirani and Totsuka =-=[10]-=- developed an interactive algorithm that finds translationally similar regions for noise removal. Often, the flawed portion is contained within a region of texture, and can be replaced by constrained ... |

16 |
An evaluation of stochastic models for analysis and synthesis of gray-scale texture
- Iversen, Lonnestad
- 1994
(Show Context)
Citation Context ...ve been proposed for texture analysis and synthesis, and an exhaustive survey is beyond the scope of this paper. We briefly review some recent and representative works and refer the reader to [8] and =-=[12]-=- for more complete surveys. Physical Simulation: It is possible to synthesize certain surface textures by directly simulating their physical generation processes. Biological patterns such as fur, scal... |

14 |
Statistical image texture analysis
- Haralick
- 1986
(Show Context)
Citation Context ...aches have been proposed for texture analysis and synthesis, and an exhaustive survey is beyond the scope of this paper. We briefly review some recent and representative works and refer the reader to =-=[8]-=- and [12] for more complete surveys. Physical Simulation: It is possible to synthesize certain surface textures by directly simulating their physical generation processes. Biological patterns such as ... |

14 | C.: Conjoint Probabilistic Subband Modeling
- POPAT
- 1997
(Show Context)
Citation Context ...etween spatial and frequency resolutions. In this paper, we choose to use the Gaussian pyramid for its simplicity and greater spatial localization (a detailed discussion of this issue can be found in =-=[19]-=-). However, other kinds of pyramids can be used instead. Neighborhood: The neighborhood can have arbitrary size and shape; the only requirement is that it contains only valid pixels. A noncausal/symme... |

10 | Deterministic texture analysis and synthesis using tree structure vector quantization
- Wei
- 1999
(Show Context)
Citation Context ... in the caption of Figure 8. a clustering probability model. Taking advantage of this clustering property, we propose to use tree-structured vector quantization (TSVQ, [7]) as the searching algorithm =-=[25]-=-. 4.1 TSVQ Acceleration Tree-structured vector quantization (TSVQ) is a common technique for data compression. It takes a set of training vectors as input, and generates a binary-tree-structured codeb... |

9 |
A context sensitive texture nib
- Malzbender, Spach
- 1993
(Show Context)
Citation Context ...ing, the algorithm can be applied to other applications, such as the image extrapolation shown in Figure 13. The algorithm could also be extended as an interactive tool for image editing or denoising =-=[15]-=-. 5.2 Temporal Texture Synthesis The low cost of our accelerated algorithm enables us to consider synthesizing textures of dimension greater than two. An example of 3D texture is a temporal texture. T... |

5 |
Vision texture. http://www-white.media.mit.edu/vismod/imagery/VisionTexture/vistex.html
- Lab
(Show Context)
Citation Context ...ction 4. 3 Synthesis Results To test the effectiveness of our approach, we have run the algorithm on many different images from standard texture sets. Figure 8 shows examples using the MIT VisTex set =-=[16]-=-, which contains real world textures photographed under natural lighting conditions. Additional texture synthesis results are available on our project website. A visual comparison of our approach with... |

4 |
Filters, random fields and maximun entropy (FRAME) - towards a unified theory for texture modeling
- Zhu, Wu, et al.
- 1998
(Show Context)
Citation Context |