[Reprinted] several hot development directions of image and graphic multimedia

Keywords: Graphic image

 

Open lab: http://openlab.net.cn/forums/437313/ShowPost.aspx

1. Digital Landscape
Digital scenery is very important in the game. Real scenarios can greatly enrich players' experience in the game process. The traditional graphic assembly line renders the surface as a polygon with texture textures. However, this method is less effective when imitating the real world surface. For this reason, most games seem a little more or less unrealistic. The latest graphics card, for example, used in Xbox, is a graphics processor (GPU) with vertex Renderer and pixel Renderer ). These GPUs make real-time rendering of realistic surfaces possible. Digital Landscape will develop a series of related technologies for game developers so that they can render realistic natural scenes in real time. In particular, it will not only provide real-time rendering engines for game developers, but also an easy-to-use Scenario Editing environment for art creators.
(1) Marine Simulation
In existing real-time applications, the ocean surface is often simulated as a plane with texture maps or concave and convex maps, with some simple light effects. The existing system can render calm ocean waves and have complex light effects. It can reach FPS on a 680 MHz Pentium iii pc equipped with a geforc4graphics card. the geometric model of a wave is represented by a dynamic displacement texture related to the angle of view, while the surface details are described by a dynamic concave-convex texture. The lighting model includes reflection and refraction, and the most important phenomenon of final, that is, the phenomenon of final water.

(2) trunk Rendering
The surface details of the trunk are the most important factors in determining its visual effects. These visual effects include complex shadows, shadows, and outlines. At present, someone has proposed a new method to build a realistic model for the surface details of the trunk, and effectively render the various visual effects mentioned above. In terms of modeling, our system is image-based: using photos of a single trunk as input, generating a height field with texture as the sample of the output trunk. With this sample, we can generate the Main Tree and the surface details of branches. In terms of rendering, a new technology, called the viewpoint-based displacement texture (vdm), represents the displacement along the angle of view. Unlike traditional displacement maps, vdm can effectively render visual effects such as shadows, shadows, and outlines without increasing the complexity of the corresponding surface geometric mesh. Vdm is based on Pixel-by-pixel processing technology. With the support of graphics hardware, it can render highly realistic object surface details in real time.

2. Facial Expression Synthesis
Through the Facial Expression Synthesis System, you only need to specify the location of the facial feature points, and the system can generate corresponding facial expressions and contain realistic details, for example, wrinkles in a face are obtained from several sample photos. The system generates facial expressions that look natural. We will also show static and continuous facial expressions, as well as 3D facial animations. By combining this system with the previous facial expression ing system, we can produce continuous facial expression sequences on any face image, which can be applied to celebrities or statues, even if they do not have an emoticon example available. Because most existing face animation systems use moving feature points to generate facial expressions, we can use our technology to improve these systems so that they can generate more realistic and natural facial expressions. Animation creators can also use our system to create any 3D facial expressions with slight skin deformation with just a few examples.

3. Visual Modeling
Each surface in the natural world has its own light reflection and penetration characteristics, so under certain lighting conditions, it will produce a unique appearance. The goal of visual modeling is to grasp the rich characteristics of various surfaces in the real world. One of the current focuses is the bidirectional texture function (BTF ). Recent work includes BTF generation on any surface, GPU-based BTF real-time rendering, and highly realistic rendering of wool objects (such as sweaters.

4. Texture Synthesis
Texture synthesis has been deeply studied in computer vision and graphics.

5. Image-based rendering
This is a promising field. Recent work includes modeling and rendering of online virtual environments, using technologies such as concentric puzzles and operations on the optical field (such as deformation of the optical field.

Graphics is still very interesting.
Due to the limitations of realistic modeling, the current trend is to combine computer vision and graphics. In Siggraph last year, singbing Kang published such a paper named the progress of the Cross-era.
I have done texture synthesis. I personally think texture synthesis has been killed. The basic entropy has been wiped out by a bunch of BT.
Image Based Rendering the technology you mentioned is also light technology.
The BTF is an article published by msra on Siggraph the year before.
We recommend that you take a look at Siggraph this year.

In terms of video encoding, I think the Research Frontier is scalable video coding, multiview video coding and high definition mosaic based video coding.
SVC is the most popular. JVT held a meeting in Hong Kong this year to discuss this topic. Currently, in H.264 extension, SVC is used. The 3D Wavelet codec of HHI and msra has many pitfalls. The relevant top-level journals are csvt

Computer Vision is no less pitfall than CG, but there are too many questions.
The mainstream journal conferences are
PAMI, cvpr, ijcv iccv (BMVC cviu ICIP is low)

In China, CG is the most optimistic. Msra Institute of computing of the Chinese Emy of Sciences Zhejiang University has many papers at the top international conferences. This year, msra of Bt published 12 articles or 17 articles. It is more than the total number of posts he has made since his arrival in China.
Video is also okay. msra and the draft offered to JVT. I remember I handed in the materials. Tsinghua South China Institute of Technology, Sun Yat-sen University, and Harbin Institute of Technology mentioned a few times. Tsinghua must have received the materials. The other few cannot be remembered. There are too many msra instances. Currently, SVC is selected by msra. Currently, a lot of PFGS (corresponding to FGS in MPEG4) is proposed by msra.
In terms of CV, this year's cvpr Tsinghua, Peking University, HKUST, MIIT, and msra are among the most popular. In particular, hkko has published 7 articles, one of which has 6 articles. Turning in is a 0 breakthrough.

The only pmai (PR, top-level journal in the CV field) in Northwest China)
It was sent by students from Xijiao zhengnanning during msra.

Although there are traps in various fields, it is difficult to dig a high level of traps. In addition, this is often the case when you compare time with each other. Your idea has just finished. The idea similar to yours has already been published at a meeting.

Research is a competition of Speed & smart.