Chinagraph’2004大会特邀报告介绍

1、Tomoyuki Nishita

Tomoyuki Nishita is a professor in the Department of Complexity Science and Engineering (also in the Department of Information Science) at University of Tokyo, Japan since 1998. He taught at Fukuyama University from 1979 to 1998 He was an associate researcher in the Engineering Computer Graphics Laboratory at Brigham Young University from 1988 to 1989. His research interests center in computer graphics including lighting models, hidden-surface removal, and antialiasing. Nishita received his BE, ME and Ph.D in Engineering in 1971, 1973, and 1985, respectively, from Hiroshima University. Dr. Nishita received Research Award on Computer Graphics from Information Processing Society of Japan in 1987. He has written twelve SIGGRAPH papers. He is one of the pioneers of Radiosity Method. He was a member of the Editorial board of the IEEE Transactions on Visualization and Computer Graphics. He has lectured at The University of Tokyo since 1994.

 

Interactive Rendering of Optical/Sound Effects due to Atmosphere and Fluid

 

                                                                                          Tomoyuki Nishita  and Yoshinori Dobashi

                                                                                      (The university of Tokyo)    (Hokkaido University)

 

In recent years, the performance of graphics hardware has made significant progress. This fact encourages researchers to study the acceleration of realistic image synthesis.  Rendering natural phenomena is one of the main topics in CG field.  Optical effects such as color of atmosphere, color of sky, shafts of lights, caustics are indispensable for the display of natural scenes.  Sound effects are also indispensable elements for the simulation of a realistic virtual environment. We would like to introduce some of interactive techniques for rendering optical effects due to atmospheric scattering and optical effects due to water.  1) We proposed an interactive rendering of the atmospheric scattering effects based on physical phenomena, that is, the phenomenon by which light is scattered by small particles in the air. This effect is the cause of the light beams produced by spotlights, shafts of light, foggy scenes, the bluish appearance of the earth’s atmosphere. In the method, look-up tables are prepared to store the intensities of the scattered light, and these are then used as textures.  2) To render the photorealistic images of the sky, the calculation taking into account multiple scattering is necessary. We proposed a method using virtual shells generated in the atmosphere. In the method, the illumination distribution of the scattering of light on the virtual shells is considered as a texture map called scattering maps.  3) We presented a method for the fast rendering of refractive and reflective caustics due to water surfaces. We render objects that are reflected and refracted due to the water surface by using reflection/refraction mapping of sampling slices.  4) We proposed a method for creating one type of sound called aerodynamic sound generated by swinging swords or by wind blowing. A major source of aerodynamic sound is vortices generated in fluids such as air.  5) In a turbulent field, the complex motion of vortices leads to the generation of sound. This type of sound is called a vortex sound. The method simulates a vortex sound by computing vorticity distributions using computational fluid dynamics.

  For another examples of interactive rendering, dynamics of hair, motion of the granular materials, and soap bubbles taking into account light interference are also discussed.

 

2. Ken Anjyo

 

Short Bio: Ken Anjyo works for OLM Digital, Inc., a digital production company in Tokyo, where he organizes the Research and Development team for making the visual effects and in-house software tools.

His research interests includes computer graphics modeling of human behaviors and natural phenomena. In particular his main concern is on construction of the mathematical and computationally tractable models. Most of his research work have been published as international journal papers or major computer graphics conference proceedings, such as SIGGRAPH and Eurographics. He has served as a paper committee member for many international computer graphics conferences, such as SIGGRAPH04, SCA04, and Pacific Graphics04.

He has also much enthusiasm in making the mathematical models usable in practice. For instance he has served as a technical director for the Pokemon Movies since 2001. Before joining OLM Digital in 2000, he was working for Hitachi, Ltd. over 17 years, where he was active as a research scientist and led the computer graphics research team. His contributions in Hitachi were dedicated not only to the commercially available software products, but also to making digital effects in several feature films, such as "Spawn, The Movie" (1997) and "Hakuchi" (1999).


Talk Title: Intuitive Flows from Mind to Images


ABSTRACT: Over recent years, the techniques for creating non-photorealistic imagery have been intensively developed centering digital cel animation and video game industry. For instance, exaggerated motion generation or cartoon shading of 3D models would be achieved by such techniques. The key issues in making these techniques practical lie in their controllability and usability. In particular, unlike a photorealistic rendering case, these non-photorealistic techniques should bring out userÅfs intention and creativity beyond physics-based reality. We therefore need a methodology or a direct, intuitive flow from what we describe in mind to the image or animation as its realization. Then we consider the following steps toward developing the techniques usable in practice:

I. What are parameterized? – Unlike a physics-based approach, we don’t have any general rule or criteria in defining parameters suitable for image generation. We must employ a different methodology according to the image generation purpose.
II. How can we make a technique endowed with the parameters? – There may exist several techniques that can provide similar results. We should select the most suitable one according to the purpose.
III. In what form can we provide the technique to a user? – This is a user interface problem. The above parameters prescribe the internal computer model so that they should not always be provided directly to the user. Rather, these should be “translated” through an intuitive interface in order that the user can easily and quickly create a desired result.


In this talk I will describe how these steps can be achieved in constructing several specific non-photorealistic techniques in a digital production workplace. The techniques include those for extracting human emotional characteristics, such as sadness or briskness, from motion-captured data, for giving (pseudo-) 3D geometry of a scene from a single 2D photograph or painting of the scene, and for creating cartoonish highlights on a 3D object used in digital cel animation. I will also address several open problems in efficiently making fake but impressive expressions needed for the entertainment industry.

 

 

3. Prof. S.Y. Shin

 

 

Sung Yong Shin is a full professor of Computer Science at the Korea Advanced Institute of Science and Technology (KAIST), Taejon, Korea. He received his BS degree in industrial engineering in 1970 from Hanyang University, Seoul, Korea, and MS and PhD degrees in industrial and operations engineering from the University of Michigan, USA, in 1983 and 1986, respectively. He has been working at KAIST since 1987. His research interests include computer graphics, real-time rendering and computational geometry. He is on the editorial boards for The Journal of Visualization and Computer Animation, Graphical Models. Prof. S.Y. Shin was the vice president of Korea Computer Graphics Society from 1993 to 1999, he is now a member of IEEE, ACM, ACM SIGGRAPH,KISS(Korea Information Science Society),KCGS (Korea Computer Graphics Society).

 

 

A Region-Based Approach to Facial Expression Cloning

 

Sung Yong Shin

KAIST, Taejon, Korea.

 

 

Human facial expressions play the most important part in delivering emotions. Thus, synthesis of convincing facial expressions has been a crucial issue in computer animation. The human process of synthesizing expressions itself is extremely complex to grasp. It involves subtle movements of facial features such as eyes, eyebrows, cheeks, the mouth, etc, driven by contraction and relaxation of muscles. Moreover, the human ability of facial expression perception is so exceptional that even a trivial displacement of a facial feature can be detected immediately. Therefore, it is difficult to synthesize facial expressions solely by simulating the complex biomechanical mechanism of a human face.

 

As animation databases have become rich, the paradigm of synthesis by reusing existing data has also been attempted in facial animation. Posed by Noh and Neumann[1], the problem of facial expression cloning is to transfer facial expressions from an existing 3D face model to a new model while preserving the emotional feelings embedded in the expressions. The original solution of the authors was based on 3D morphing to apply the deformed motion vectors of the source model to the target model. Later, Pyun et al.[2] proposed a new method based on scattered data  interpolation [3, 4, 5, 2] and demonstrated its effectiveness in both quality and efficiency.

 

In the approach of Pyun et al., the animator provides target key-models, each of which corresponds to a source key-model, to realize his/her imagination. Given two or more sets of face key-models, each set representing a different aspect of facial expressions, their approach enumerates all combinations of key-models from different sets. For example, given 6 key-models for emotional states and 13 key-models for verbal communications, 78(6 ´ 13) key-models need to be created to reflect both emotional and verbal expressions together[2]. In general, the number of face key-models grows combinatorially as the number of aspects increases. Moreover, merging two or more key-models into one is a non-trivial task. Yet another weakness is that the key-models are implicitly assumed to be symmetrical, that is, the left and right halves of every key-model are symmetrical, which restricts the domain of applications.

 

In this talk, we present a region-based method for facial expression cloning. Our approach is based on results in psychology[6]: A face can be split into several regions that behave as coherent units to exhibit facial expressions. For example, the region containing an eye is used for emotional expressions. On the other hand, that containing the mouth is mainly for verbal expressions. Our objective is to remedy the weakness of the previous method[2] by directly dealing with those regions containing facial features while still preserving its advantage, that is, on-line, real-time performance. Consequently, more dynamic expressions including asymmetric ones can be synthesized with a small set of key-models. Because of supreme time efficiency and compact key-model representations, our approach could be applicable for applications on the web or mobile platforms[7].

 

To achieve the objective, we need not only a good segmentation tool but also an effective way to combine separately-cloned features seamlessly. We address those issues by learning the contribution of every facial feature to the movement of a vertex from user-supplied face key-models. Moreover, the animators can breathe their creativity and imagination into the target key-models while preparing them.

 

Adopting the framework of scattered data interpolation, our method consists of two parts: analysis of face key-models and synthesis of facial expressions, as illustrated in the below figure.

 

Figure : The overall structure of our method

 

The analysis part is the preprocessing step consisting of two tasks: region segmentation and parameterization. Given source key-models and their corresponding target key-models, we first segment every key-model into three regions, each containing one of the chosen facial features, that is, the left eye, the right eye, and the mouth. These regions may overlap each other near their boundaries. As the results of region segmentation, we obtain three sets of key-shapes(features) for the source key-models and their corresponding sets for the target key-models. We apply principal component analysis (PCA) to the source key-shapes of each set to place the corresponding target key-shapes in their parameter space.

 

The synthesis part is the actual cloning step consisting of three tasks: parameter extraction, key-shape blending, and region composition. The first two tasks are typical operations for the scattered data interpolation based on cardinal basis functions[3, 4, 5, 2], which are performed frame by frame on every feature of a input face model separately, to obtain the corresponding feature of the output face model. The final task is to composite the three separately-blended features to produce the output face model in each frame.

 

 

[1] J. Y. Noh and U. Neumann. Expression cloning. In Proc. SIGGRAPH’01, pages 277-288, 2001.

[2] H. Pyun, Y. Kim, W. Chae, H. Y. Kang, and S. Y. Shin. An example-based approach

for facial expression cloning. ACM SIGGRAPH/EUROGRAPHICS Symposium on

Computer Animation, pages 167-176, 2003.

[3] C. Rose, M. F. Cohen, and B. Bodenheimer. Verbs and adverbs: Multidimensional

motion interpolation. IEEE Computer Graphics and Applications, 18(5):32-40, Sept. 1998.

[4] P. P. Sloan, C. F. Rose, and M. F. Cohen. Shape by example. 2001 Symposium on Interactive 3D Graphics, pages 135-144, 2001.

[5] S. I. Park, H. J. Shin, and S. Y. Shin. On-line locomotion generation based on motion blending. ACM SIGGRAPH Symposium on Computer Animation, pages 105-111, 2002.

[6] P. Ekman and W. V. Friesen. Unmasking the face: A guide to recognizing emotions from facial clues. Prentice-Hall Inc. Englewood Cliffs, New Jersey, 1975.

[7] I. S. Pandzic and R. Forchheimer. MPEG-4 Facial Animation. Wiley, 2002.

 

 

 

4. Prof. Baining Guo

Baining Guo is the research manager of the Internet Graphics group at Microsoft Research Asia (formerly Microsoft Research China). Before joining Microsoft, Baining was a senior staff researcher in the Microcomputer Research Labs at Intel Corporation in Santa Clara, California, where he worked on graphics architecture. Before moving to the Silicon Valley, Baining worked at University of Colorado, University of Toronto, and York University. He was also a visiting professor at Ecole Nationale Superieure Des Telecommunications and Princeton University. Baining received his Ph.D. and M.S. from Cornell University and his B.S. from Beijing University. His research interests are mainly in modeling and rendering areas, including texture synthesis, reflectance and shading models, real-time rendering, natural phenomena, and geometry modeling. Baining is currently an Associate Editor of IEEE Trans. on Visualization and Computer Graphics.

 

3D modeling and rendering techniques for games and cool technologies

 

Baining Guo

Microsoft Research Asia

 

Abstact:

This talk will provide an overview of some of our projects in the areas of 3D modeling and rendering techniques for games and cool technologies for interacting with digital images. The 3D modeling and rendering techniques include generalized displacement mapping, shell texture, Poisson mesh, iso-charts (for generating texture atlas), tree rendering, etc. Techniques for interacting with images include lazy snapping (an image cut-out tool), Poisson matting, video toning (for stylizing your home videos) and more. We will both describe the techniques and provide brief demos of each. Part of the talk will provide an overview of five presentations we gave at Siggraph.