high-quality video view interpolation larry zitnick interactive visual media group microsoft...

65
High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Post on 20-Dec-2015

215 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

High-Quality Video View Interpolation

Larry ZitnickInteractive Visual Media Group

Microsoft Research

Page 2: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

3D video

Lumigraph Light field

Geometry centric Image centric

Warping InterpolationPolygon rendering + texture mapping

Fixed geometry

View-dependent geometry

View-dependent texture

Sprites with depth

Layered depth Image

Page 3: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Current practice

Many cameras

Motion Jitter

vs.

free viewpoint video

Page 4: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Current practice

Many cameras

Motion Jitter

vs.

free viewpoint video

Page 5: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Video view interpolation

Fewer cameras

Smooth Motion

Automatic

and

Real-time rendering

Page 6: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

System overview

OFFLINE

ONLINE

Video Capture

Stereo Compression

SelectiveDecompression

Render

File

Representation

Video Capture

Page 7: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

concentratorsconcentrators

hard diskshard disks controlling

laptopcontrollinglaptop

camerascamerascameras

Page 8: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Calibration

Zhengyou Zhang, 2000

Page 9: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Input videos

Page 10: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

System overview

OFFLINE

ONLINE

Video Capture

Stereo Compression

SelectiveDecompression

Render

File

Representation

Video Capture

Stereo

Page 11: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Key to view interpolation: Geometry

Stereo Geometry

Camera 1 Camera 2

Image 1 Image 2

Virtual Camera

Page 12: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Match ScoreMatch ScoreMatch Score Match Score

Good

Bad

Image correspondence

Correct

Image 1 Image 2

Leg

Wall

Incorrect

Page 13: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Why segments?

Better delineation of boundaries.

Page 14: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Why segments?

Larger support for matching.

Handle gain and offset differences without global model (Kim, Kolmogorov and Zabih, 2003.)

Page 15: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Why segments?

More efficient.

786,432 pixels vs. 1000 segments

Compute disparities per segment rather than per pixel.

Page 16: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Segmentation

Many methods will work:

Graph-based (Felzenszwalb and Huttenlocher, 2004)

Mean Shift (Comaniciu, et al. 2001)

Min-cut (Boykov et al. 2001)

Others…

Page 17: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Segmentation: Important properties

Not too large, not too small…

As large as possible while not spanning multiple objects.

Page 18: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Segmentation: Important properties

Stable Regions

Page 19: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Segmentation: Our Approach

First average…

…then segment.

Anisotropic smoothing

Page 20: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Segmentation: Result

Close-up

Page 21: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Matching segments

Many measures will work: SSD Normalized correlation Mutual information

Depends on color balancing and image quality.

Page 22: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Matching segments: Important properties

Never remove correct matches.

Remove as many false matches as possible

Use global methods to remove remaining false positives.

Page 23: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Matching segments: Our approach

Create gain histogram

Good match

Bad match0.8

0.8 1.25

1.25

1

0

p

pgain

Page 24: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Image 1 Image 2

Local matching

Low textureLow texture

Page 25: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Number of states = number of depth levels

Image 2Image 1

Global regularization

Create MRF (Markov Random Field):

A F

E

DC

B

Each segment is a node

P Q R

S T

U

Page 26: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Global regularization

Disparity Images

Likelihood (data term)

Prior (regularization term)

Page 27: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Image 2Image 1

Global regularization

A F

E

DC

B

colorA ≈ colorB → zA ≈ zB

P Q R

S T

U

Page 28: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Global regularization

i iSk knl

kliliki ddΝDP 2,;

Variance – % of border and similarity of colorNormal distribution

A F

E

DC

B

A F

E

DC

B

A F

E

DC

B

Disparity Disparity Disparity

Page 29: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Multiple disparity maps

Compute a disparity map for each image.

We want the disparity maps to be consistent across images…

Page 30: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Image 2Image 1

Consistent disparities

A F

E

DC

B

zA ≈ zP, zQ, zS

P Q R

S T

U

A

A

Page 31: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Consistent disparities

Disparities dependent on neighboring disparities.

Likelihood includes neighboring disparities.

Page 32: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Consistent disparities

if not occluded

if occluded

Use original data term if not occluded.

Bias disparities to lie behind known surfaces when occluded.

Page 33: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Is the segment occluded?

Not occludedOccluded

Ii

Page 34: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

If occluded…

Occluded

Disparity

Ii

Page 35: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Iteratively solve MRF

Page 36: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Depth through time

Page 37: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

MattingInterpolated view without matting

Background Surface

Foreground Surface

Camera

Foreground Alpha

Background

Bayesian MattingChuang et al. 2001

Strip Width

Background

Foreground

Page 38: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Rendering with matting

MattingNo Matting

Page 39: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

System overview

OFFLINE

ONLINE

Video Capture

Stereo Compression

SelectiveDecompression

Render

File

RepresentationStereo Representation

Page 40: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Representation

Main Layer:

Color

Depth

Main

Boundary

Boundary Layer:

Color

Depth

Alpha

Background

ForegroundStrip Width

Page 41: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

System overview

OFFLINE

ONLINE

Video Capture

Stereo Compression

SelectiveDecompression

Render

File

RepresentationRepresentation Compression

Page 42: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Time = 0 Time = 1

Camera 1

Camera 2

Camera 3

Camera 4

Compression

Page 43: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Time = 0 Time = 1

Camera 1

Camera 2

Camera 3

Camera 4

TemporalPrediction

Compression

Page 44: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Time = 0 Time = 1

Camera 1

Camera 2

Camera 3

Camera 4

SpatialPrediction

Compression

Page 45: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Spatial prediction

Depth and Texture

ReferenceCamera

PredictedCamera

Page 46: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

ReferenceCamera

PredictedCamera

Depth and Texture

Warped

Spatial prediction

Page 47: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Error Signal

ReferenceCamera

PredictedCamera

Warped Depth and Texture

Spatial prediction

_

+

Page 48: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Reconstructed (after error signal is added)

ReferenceCamera

PredictedCamera

Warped Depth and Texture

Spatial prediction

Page 49: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research
Page 50: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Boundary layer coding

Depth Color Texture Alpha Matte

Color

Depth

Alpha

Use our own shape coding method similar to MPEG-4Use our own shape coding method similar to MPEG-4

Page 51: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

System overview

OFFLINE

ONLINE

Video Capture

Stereo Compression

SelectiveDecompression

Render

File

Representation

SelectiveDecompression

Render

Compression

File

Page 52: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Rendering

Source Cameras

Page 53: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

ProjectMain Layer

ProjectBoundary Layer

Composite

ProjectMain Layer

ProjectBoundary Layer

Rendering

Page 54: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

ProjectedVideo of background

depth

Video of background

color

Depth Color

GPU

Vertex Shader Pixel ShaderPosition,Texture Coord

Rendering the main layer (Step 1)

Z-Buffer

Color Buffer

Page 55: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research
Page 56: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research
Page 57: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Main Layer Depth

Projected

GPU

Pixel Shader

CPU

Generate Erase Mesh

Rendering the main layer (Step 2)

Color Buffer

Z-Buffer

Locate Depth Discontinuities

Page 58: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

GPU

Compositing

CPU

Generate Boundary Mesh

Boundary Depth

Boundary RGBA

Rendering boundary layerProjected Main Layer

Vertex Colors Color Buffer

Z-Buffer

Projected

Page 59: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Graphics for Vision

Use the GPU for vision.

Real-time stereo – (Yang and Pollefeys, CVPR 03)

Page 60: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

ProjectMain Layer

ProjectBoundary Layer

Composite

ProjectMain Layer

ProjectBoundary Layer

Rendering

Page 61: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

GPU

Pixel Shader

Camera 1 Camera 2

Final Result

Final composite

Weights based on proximity to virtual viewpoint

Compositing views

Page 62: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

DEMO

Page 63: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

“Massive Arabesque” videoclip

Page 64: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Future work

Mesh simplification More complicated scenes Temporal interpolation (use optical flow) Wider range of virtual motion 2D grid of cameras

Page 65: High-Quality Video View Interpolation Larry Zitnick Interactive Visual Media Group Microsoft Research

Summary

Sparse camera configuration High-quality depth recovery Automatic matting New two-layer representation Inter-camera compression Real-time rendering