영상 기반 표현 (image Based Representation)

중앙대학교 CG Lab.

류승택2000. 9. 14

Introduction

3D model-based Rendering Explicit use of 3

Dimensional models Uses conventional

rendering pipeline Speed dependent on

scene complexity Relies on hardware

accelerator for speed Requires sophisticated

software for realism

Image-based Rendering Directly uses collection of

image Based on interpolation or

pixel reprojection Speed independent on

scene complexity Relies on processor

speed Realism depends on input

images

Traditional vs Image-based Representation

Traditional Modeling and Rendering

Image based Modeling and Rendering

영상 기반 분류

영상기반 표현 방법 (Image Based Representation) 영상기반 모델링 (Image Based Modelling)

• 2 차원 영상으로부터 3 차원 모델을 복원 및 표현 영상기반 렌더링 (Image Based Rendering)

• 2 차원 영상들의 조합으로부터 새로운 시점의 영상 생성 영상기반 가속화 (Image Based Acceleration)

• 가까운 거리에서는 실제 Geometry Model 사용하고 먼 거리에 있는 물체에 대해 Image 로 대치 시킴

영상기반 라이팅 (Image Based Lighting)• 2 차원 영상으로부터 조명 정보를 추출하여 새로운 광원을 부여

Image Based Modeling:

The Chevette Project

Image Based Modeling: The Campanile Movie

Overview 사진으로부터 모델을 복원 2 차원 이미지로부터 모델의 3 차원 공간정보획득

Image Based Rendering: The Plenoptic Function

Plenoptic Function The light in a scene can be

described by a multi-dimensional function

infeasible since it would require measurements from all points of the scene and in all direction at all times

Image Based Rendering: Environment Map

Mosaics (stitching) combination of at least two different images yield a higher resolution or larger image

Environment Map reduce the dimensionality of the plenoptic function by restric

ting the observer to one point

Image Based Rendering: Light Field & Lumigraph

Light Slab Two-plane Parameterization

• reducing the plenoptic function into a four-dimensional table

• Light flows from uv plane to st plane

• Planes in any position

Image Based Rendering: View Morphing

View Morphing Techniques for combining image Three steps

• Prewarp Morph Postwarp

Image Based Rendering: Layered Depth Image

Sprite with Depth Layered Depth Image

Image Based Acceleration: Realtime Walkthru

가상공간의 실시간 탐색을 위한 연구 가상공간을 높은 화질로 표현하기 위해서는 복잡하고 많은

양의 모델이 필요 실시간 탐색이 어려움

• 보이는 물체만을 렌더링 (Visibility Culling)

• 가까운 거리에서는 실제 Geometry Model 사용• 먼거리에 있는 물체에 대해

– LOD(Level Of Detail) 를 적용하거나– Image 로 대치 시킴 (Image Based Representation)

Image Based Acceleration: Image Based Primitives

Hybrid Approach Sprite Imposter Nailboards Portal Texture image caches

<Image Based Primitive>

Image Based Lighting: Global Illumination

Others:

Tour into the picture

Others:

Multiple Center of Projection Images

Overview Strip 카메라로부터 획득한 MCOP Image 를 이용하여 영상표현

Acquiring MCOP Image

Others:

Making face

Art:

Virtual Camera

Array of pictures taken onto same roll of film Playback achieves “frozen time” effect

Art:

The Matrix

Applied Techniques photogrammetric modeling, projective texture-mapping Also applied in shots of the Tom Cruise film Mission Impossible II.

Graphics and Vision Technique

Appearance based

Physicallybased

Lumigraphand

light field

Image morphing

Sprites

Planarsprites

Spriteswith depth

Layereddepthimage

Triangle scan conversion

View-dependenttextures

Texture mapping

Geometriclevel

of detail

Global illumination

Radiosity Monte Carloray

tracing

image Geometric model

Graphics

MultiviewStereo

Optical flowestimation

Plane-sweepstereo

Layeredstereo

Depth maprecovery

Texturerecovery

Imagemosaics

Range datamerging

Curves to3D mesh

Silhouettesto volume

Geometryfitting

Illuminationestimation

Reflectionestimation

Vision

참고문헌

Mcmillan, L. and Bishop, G. "Plenoptic modeling : An image-based rendering rendering system." SIGGRAPH '95 PROCEEDING, 1995, pp 39--46.

Anders Gustafsson and Henrik Turbell, Image-Based Rendering, Linkoping University, Sweden, 1997

Sing Bing Kang, A Survey of Image-based Rendering Techniques, Technical Report, CRL 97/4, Digital Equipment Corp., Cambridge Research Lab, Cambridge, Mass., 1997.

Leonard McMillian Jr, An Image-based Approach to three-dimensional computer graphics, PH.D. DISSERTATION, UNC-CH DEPT. OF COMPUTER SCIENCE, APRIL 1997. (PUBLISHED AS UNC COMPUTER SCIENCE TECHNICAL REPORT TR97-013)

Christoph Bregler, Michael F. Cohen, Paul Debevec, Leonard McMillan, Francois X. Sillion, Richard Szeliski, Image-based Modeling, Rendering and Lighting, SIGGRAPH 99 Course note, 1999

Visual Navigation of Large Environments Using Texture Clusters, Paulo, 93 Dynamic Generated Imposter, Gernot Schaufler, 95 Efficient Imposter Manipulation for Real-Time Visualization of Urban Scenery, Francois Sillion, EG’ 9

7 Multi-Layered imposters for accelerated rendering, EG 99 “Architectural Walkthroughs Using Portal Textures”, 1997 년 (IEEE Visualization 97) “3D Image Warping in Architectural Walkthroughs”, 1998. 3 (VRAIS ‘98) “Efficient Warping for Architectural Walkthroughs using LDIs”, 1998. 10 (IEEE Visualization 98) “Image for Accelerating Architectural Walkthroughs”, 1998. 12 (CG&A)