## Texture Mixing and Texture Movie Synthesis using Statistical Learning (2001)

### Download From

IEEE### Download Links

- [www.cs.huji.ac.il]
- [www.wisdom.weizmann.ac.il]
- [www.cs.technion.ac.il]
- [www.cs.huji.ac.il]
- [www.cs.huji.ac.il]
- DBLP

### Other Repositories/Bibliography

Venue: | IEEE Transactions on Visualization and Computer Graphics |

Citations: | 114 - 7 self |

### BibTeX

@ARTICLE{Bar-joseph01texturemixing,

author = {Ziv Bar-joseph and Ran El-Yaniv and Dani Lischinski and Michael Werman},

title = {Texture Mixing and Texture Movie Synthesis using Statistical Learning},

journal = {IEEE Transactions on Visualization and Computer Graphics},

year = {2001},

volume = {7},

pages = {120--135}

}

### Years of Citing Articles

### OpenURL

### Abstract

We present an algorithm based on statistical learning for synthesizing static and time-varying textures matching the appearance of an input texture. Our algorithm is general and automatic, and it works well on various types of textures including 1D sound textures, 2D texture images and 3D texture movies. The same method is also used to generate 2D texture mixtures that simultaneously capture the appearance of a number of different input textures. In our approach, input textures are treated as sample signals generated by a stochastic process. We first construct a tree representing a hierarchical multi-scale transform of the signal using wavelets. From this tree, new random trees are generated by learning and sampling the conditional probabilities of the paths in the original tree. Transformation of these random trees back into signals results in new random textures. In the case of 2D texture synthesis our algorithm produces results that are generally as good or better than those produce...

### Citations

2153 |
Time Series Analysis for Forecasting and Control
- Box, Jenkins
- 1970
(Show Context)
Citation Context ...ntations are very common for modeling 1D and 2D signals and there are various, well established approaches to estimating the underlying unknown source with 5srespect to such representations (see e.g. =-=[4, 20]). A-=-lthough such representations have been successfully used in many application domains, they are not adequate for representation and analysis of “textural signals” such as the ones treated in this p... |

1767 | Biorthogonal bases of compactly supported wavelets
- Cohen, A, et al.
- 1992
(Show Context)
Citation Context ...he signal, with the wavelet basis functions of a certain resolution (at all possible translations of the basis functions). In this work we use two different types of wavelets: the Daubechies wavelets =-=[7] a-=-nd the steerable pyramid [25]. The steerable pyramid transform is used to analyze 2D texture images and the spatial dimensions of texture movies. Daubechies wavelets are used to analyze the TVT’s te... |

573 |
An image synthesizer
- Perlin
- 1985
(Show Context)
Citation Context ...ost previous work in texture synthesis has focused on the development of procedural textures, where complex interesting patterns are produced by a program executed before or during the shading process=-=[14, 22, 30, 31, 32]-=-. Although some of the most compelling synthetic imagery has been produced with the aid of procedural textures, the disadvantage of this approach is that it can be difficult to control and/or predict ... |

467 | Shiftable multiscale transforms
- Simoncelli, Freeman, et al.
- 1992
(Show Context)
Citation Context ...sis functions of a certain resolution (at all possible translations of the basis functions). In this work we use two different types of wavelets: the Daubechies wavelets [7] and the steerable pyramid =-=[25]. -=-The steerable pyramid transform is used to analyze 2D texture images and the spatial dimensions of texture movies. Daubechies wavelets are used to analyze the TVT’s temporal dimension. Thus, for sou... |

400 | Pyramid-based texture analysis/synthesis
- Heeger, Bergen
- 1995
(Show Context)
Citation Context ...mputer graphics. Recently, several techniques have emerged in the computer graphics literature that are able to analyze an input texture sample and synthesize many new random similar-looking textures =-=[8, 18, 34]. -=-This work extends these techniques in several important ways. First, we describe a new texture mixing algorithm — a statistical learning algorithm that operates on several different input texture sa... |

347 | Particle systems - technique for modeling a class of fuzzy objects
- Reeves
- 1983
(Show Context)
Citation Context ...mulations. For example, steam, fog, smoke, and fire have been simulated in this manner [15, 13, 27, 28]. Explosions, fire, and waterfalls have been successfully simulated by animated particle systems =-=[23, 24, 26]-=-. Simplified physically-based models have also been used to produce synthetic waves and surf [17, 21]. While the techniques mentioned above have been able to generate impressive results of compelling ... |

327 |
Image Compression through Wavelet Transfrom Coding
- Devore, Jawerth, et al.
- 1992
(Show Context)
Citation Context ...ssess the size of the temporal features in the sequence simply by observing it. Our technique for choosing a threshold for the temporal dimension is inspired by wavelet compression methods for images =-=[12]-=-. The idea behind wavelet compression is to zero out coefficients with L1 norm less than some small number a. This decimation of the coefficients results in little perceptual effect on subjective imag... |

288 | A Multiresolution Spline With Application to Image Mosaics
- Burt, Andelson
- 1983
(Show Context)
Citation Context ...me-varying textures. Texture mixing, a process of generating a new texture that contains features from several different input textures, has been addressed in several previous works. Burt and Adelson =-=[5]-=- produce smooth transitions between different textures by weighted averaging of the Laplacian pyramid coefficients of the textures. This technique is very effective for seamless image mosaicing, but i... |

249 | Elements of information theory. Wiley series in telecommunications - Cover, Thomas - 1991 |

249 |
Wavelets for computer graphics: theory and applications
- Stollnitz, Derose, et al.
- 1996
(Show Context)
Citation Context ...l transform. Wavelets have been effectively utilized in many different fields. A comprehensive review of wavelets applied to computer graphics can be found in a book by Stollnitz, Salesin, and DeRose =-=[29]-=-. A wavelet representation is a multi-scale decomposition of the signal and can be viewed as a complete tree, where each level stores the projections of the signal, with the wavelet basis functions of... |

243 | Multiresolution sampling procedure for analysis and synthesis of texture image
- Bonet
- 1997
(Show Context)
Citation Context ...mputer graphics. Recently, several techniques have emerged in the computer graphics literature that are able to analyze an input texture sample and synthesize many new random similar-looking textures =-=[8, 18, 34]. -=-This work extends these techniques in several important ways. First, we describe a new texture mixing algorithm — a statistical learning algorithm that operates on several different input texture sa... |

242 | Generating Textures on Arbitrary Surfaces Using Reaction-Diffusion
- Turk
- 1991
(Show Context)
Citation Context ...ost previous work in texture synthesis has focused on the development of procedural textures, where complex interesting patterns are produced by a program executed before or during the shading process=-=[14, 22, 30, 31, 32]-=-. Although some of the most compelling synthetic imagery has been produced with the aid of procedural textures, the disadvantage of this approach is that it can be difficult to control and/or predict ... |

213 |
Texturing and Modeling: A Procedural Approach
- Ebert, ed
- 1994
(Show Context)
Citation Context ...ost previous work in texture synthesis has focused on the development of procedural textures, where complex interesting patterns are produced by a program executed before or during the shading process=-=[14, 22, 30, 31, 32]-=-. Although some of the most compelling synthetic imagery has been produced with the aid of procedural textures, the disadvantage of this approach is that it can be difficult to control and/or predict ... |

205 | Filters, random fields and maximum entropy (FRAME) – towards the unified theory for texture modeling
- Zhu, Wu, et al.
- 1996
(Show Context)
Citation Context ...mputer graphics. Recently, several techniques have emerged in the computer graphics literature that are able to analyze an input texture sample and synthesize many new random similar-looking textures =-=[8, 18, 34]. -=-This work extends these techniques in several important ways. First, we describe a new texture mixing algorithm — a statistical learning algorithm that operates on several different input texture sa... |

168 |
Approximate and Probabilistic Algorithms for Shading and Rendering Structured Particle Systems
- Reeves, Blau
- 1985
(Show Context)
Citation Context ...mulations. For example, steam, fog, smoke, and fire have been simulated in this manner [15, 13, 27, 28]. Explosions, fire, and waterfalls have been successfully simulated by animated particle systems =-=[23, 24, 26]-=-. Simplified physically-based models have also been used to produce synthetic waves and surf [17, 21]. While the techniques mentioned above have been able to generate impressive results of compelling ... |

137 | Universal Prediction
- Merhav, Feder
- 1998
(Show Context)
Citation Context ...ntations are very common for modeling 1D and 2D signals and there are various, well established approaches to estimating the underlying unknown source with 5srespect to such representations (see e.g. =-=[4, 20]). A-=-lthough such representations have been successfully used in many application domains, they are not adequate for representation and analysis of “textural signals” such as the ones treated in this p... |

133 | Reaction-diffusion textures
- Witkin, Kass
- 1991
(Show Context)
Citation Context |

132 | Depicting fire and other gaseous phenomena using diffusion processes
- Stam, Fiume
- 1995
(Show Context)
Citation Context ...esis of dynamic natural phenomena has mostly been possible only via computationally intensive physically based simulations. For example, steam, fog, smoke, and fire have been simulated in this manner =-=[15, 13, 27, 28]-=-. Explosions, fire, and waterfalls have been successfully simulated by animated particle systems [23, 24, 26]. Simplified physically-based models have also been used to produce synthetic waves and sur... |

128 |
A Simple Model of Ocean Waves
- Fournier, Reeves
- 1986
(Show Context)
Citation Context ...Explosions, fire, and waterfalls have been successfully simulated by animated particle systems [23, 24, 26]. Simplified physically-based models have also been used to produce synthetic waves and surf =-=[17, 21]-=-. While the techniques mentioned above have been able to generate impressive results of compelling realism, a custom-tailored model must be developed for each type of simulated phenomena. Furthermore,... |

112 | Turbulent wind fields for gaseous phenomena
- Stam, Fiume
- 1993
(Show Context)
Citation Context ...esis of dynamic natural phenomena has mostly been possible only via computationally intensive physically based simulations. For example, steam, fog, smoke, and fire have been simulated in this manner =-=[15, 13, 27, 28]-=-. Explosions, fire, and waterfalls have been successfully simulated by animated particle systems [23, 24, 26]. Simplified physically-based models have also been used to produce synthetic waves and sur... |

111 | Particle Animation and Rendering Using Data Parallel Computation
- Sims
- 1990
(Show Context)
Citation Context ...mulations. For example, steam, fog, smoke, and fire have been simulated in this manner [15, 13, 27, 28]. Explosions, fire, and waterfalls have been successfully simulated by animated particle systems =-=[23, 24, 26]-=-. Simplified physically-based models have also been used to produce synthetic waves and surf [17, 21]. While the techniques mentioned above have been able to generate impressive results of compelling ... |

99 |
Cellular texture basis function
- Worley
- 1996
(Show Context)
Citation Context |

96 | Modeling and estimation of multiresolution stochastic processes
- Basseville, Benveniste, et al.
- 1992
(Show Context)
Citation Context ... at different scales and locations. Many recent research efforts suggest that a better representation for such signals are multi-resolution structures, such as wavelet-based representations (see e.g. =-=[2, 33]-=-). Basseville et al.[3, 2] provide a theoretical formulation of statistical modeling of signals by multi-resolution structures. In particular, these results consider a generative stochastic source (th... |

92 | Rendering and animation of gaseous phenomena by combining fast volume and scanline A-buffer techniques
- Ebert, Parent
(Show Context)
Citation Context ...esis of dynamic natural phenomena has mostly been possible only via computationally intensive physically based simulations. For example, steam, fog, smoke, and fire have been simulated in this manner =-=[15, 13, 27, 28]-=-. Explosions, fire, and waterfalls have been successfully simulated by animated particle systems [23, 24, 26]. Simplified physically-based models have also been used to produce synthetic waves and sur... |

74 | A non-parametric multiscale statistical model for natural images
- Bonet, Viola
- 1997
(Show Context)
Citation Context ... such a technique. For example, methods for statistical learning of 2D texture images have been successfully applied not only to texture synthesis, but also to texture recognition and image denoising =-=[10]-=-. These applications are made possible by realizing that statistical learning of 2D textures implicitly constructs a statistical model describing images of a particular class. Similarly, our approach ... |

72 |
Modeling waves and surf
- PEACHEY
- 1986
(Show Context)
Citation Context ...Explosions, fire, and waterfalls have been successfully simulated by animated particle systems [23, 24, 26]. Simplified physically-based models have also been used to produce synthetic waves and surf =-=[17, 21]-=-. While the techniques mentioned above have been able to generate impressive results of compelling realism, a custom-tailored model must be developed for each type of simulated phenomena. Furthermore,... |

34 |
Wavelet-based representations for a class of selfsimilar signals with application to fractal modulation
- Wornell, Oppenheim
- 1992
(Show Context)
Citation Context ... at different scales and locations. Many recent research efforts suggest that a better representation for such signals are multi-resolution structures, such as wavelet-based representations (see e.g. =-=[2, 33]-=-). Basseville et al.[3, 2] provide a theoretical formulation of statistical modeling of signals by multi-resolution structures. In particular, these results consider a generative stochastic source (th... |

33 | Agnostic classification of Markovian sequences
- El-Yaniv, Fine, et al.
- 1997
(Show Context)
Citation Context ...stical sources of sequences, which are simply paths in the representing tree. The particular estimation algorithm for sequences we chose to use is an extension of the algorithm due to El-Yaniv et al. =-=[16]-=-, which operates on sequences over a finite alphabet, rather than on real numbers. Given a sample sequence S, this algorithm generates new random sequences which could have been generated from the sou... |

26 |
Solid Spaces and Inverse Particle Systems for Controlling the Animation of Gases and Fluids, The Visual Comp
- Ebert, Carlson, et al.
- 1791
(Show Context)
Citation Context |

21 |
Multiscale Autoregressive Processes, Parts I and II
- Basseville, Benveniste, et al.
- 1992
(Show Context)
Citation Context ...cations. Many recent research efforts suggest that a better representation for such signals are multi-resolution structures, such as wavelet-based representations (see e.g. [2, 33]). Basseville et al.=-=[3, 2]-=- provide a theoretical formulation of statistical modeling of signals by multi-resolution structures. In particular, these results consider a generative stochastic source (that can randomly generate m... |

16 |
Texture recognition using a non-parametric multi-scale statistical model
- JS, Viola
- 1998
(Show Context)
Citation Context ...hesize mixtures of different textures, as illustrated in Section 5.2. It should be noted that De Bonet and Viola briefly mention the possibility of extending their approach to multiple input examples =-=[11]-=-. A second difference is that when our algorithm generates level i tree nodes, we are actually looking at nodes in level is1. For each such node x we are looking for nodes in the analysis pyramids (in... |

8 |
Novel statistical multiresolution techniques for image synthesis, discrimination, and recognition
- Bonet
- 1997
(Show Context)
Citation Context ...from other less structured textures. 5.3 Comparison With De Bonet’s Algorithm As mentioned earlier the application of our algorithm to the task of 2D texture synthesis is similar to De Bonet’s met=-=hod [9]-=-. However, there are in fact several important differences between the two approaches. After explicitly stating our algorithm, we can now explain those differences in detail. The most important distin... |

2 |
Statistical learning of granular synthesis parameters with applications for sound texture synthesis
- Bar-Joseph, Dubnov, et al.
- 1999
(Show Context)
Citation Context ...et they manage to capture the essential perceptual characteristics and the global temporal behavior observed in the input sequence. A specialized version of this method, described in a separate paper =-=[1]-=-, is able to synthesize 1D sound textures, such as sounds of traffic, water, etc. The natural applications of texture movie synthesis are in special effects for motion pictures and television, compute... |