samuel w. hasinoff sing bing kang richard szeliski interactive visual media group microsoft research...

23
Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer Science University of Toronto [email protected] Boundary Matting for View Synthesis 2 nd Workshop on Image and Video Registration, July 2, 2004

Post on 15-Jan-2016

215 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Samuel W. Hasinoff Sing Bing Kang Richard Szeliski

Interactive Visual Media GroupMicrosoft Research{sbkang,szeliski}@microsoft.com

Dept. of Computer ScienceUniversity of [email protected]

Boundary Matting for View Synthesis

2nd Workshop on Image and Video Registration, July 2, 2004

Page 2: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

MotivationSuperior view synthesis & 3D editing from N-view stereo

Key approach: occlusion boundaries as 3D curves

• More suitable for view synthesis• Boundaries estimated to sub-pixel

Two major limitations – even with perfect stereo!• Resampling blur• Boundary artifacts

Page 3: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

B2B3

Matting problem: Unmix the foreground & background

Matting from Stereo

BFC )1(

Triangulation matting (Smith & Blinn, 1996)

• multiple backgrounds• fixed viewpoint & object

F

B1

Extension to stereo• Lambertian assumption

F

B3B1 B2

underdetermined

Page 4: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Occlusion Boundaries in 3D Model boundaries as 3D splines (currently linear) Assumptions

boundaries are relatively sharp relatively large-scale objects no internal transparency

view 1 view 3view 2 (reference)

3D world

Page 5: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Geometric View of Alpha

alpha partial pixel coverage on F side

simulate blurring by convolving with 2D Gaussian

otherwise,0

0)(,1)(

xdx

),0()(),( Gxx

j

j x)(

alpha depends only on projected 3D curve, x

integration over each pixel

F B

pixel j

Page 6: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Related Work Natural image matting [Chuang et al., 2001]

based on color statistics

Intelligent scissors [Mortenson, 2000]

geometric view of alpha

- single image- user-assisted

Page 7: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Related Work Bayesian Layer estimation [Wexler and Fitzgibbon, 2002]

matting from multiple images using triangulation + priors

- requires very high-quality stereo- alpha calculated at pixel level, only for reference - not suitable for view synthesis

Page 8: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Boundary Matting Algorithm

3D world

view 1 view 3view 2 (reference)

find occlusion boundary in reference view backproject to 3D using stereo depth project to other views initial guess for Bi and F optimize matting

optimize

Page 9: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Initial Boundaries From Stereo Find depth discontinuities Greedily segment longest four-connected curves

Spline control points evenly spaced along curve

Tweak - snap to strongest nearby edge

Page 10: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Background Estimation

F

B1 B2

Use stereo to grab corresponding background-depth pixels from nearby views (if possible)

Color consistency check to avoid mixed pixels

B3

occluded

Page 11: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Foreground Estimation

Invert matting equation, given 3D curve and B

Aggregate F estimates over all views

viewsviews

ii

iii FF

1

2

1

2 )(ˆˆ

BCF )1()(ˆ

BFC )1(

Page 12: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Optimization

Objective: Minimize inconsistency with matting

over curve parameters, x, and foreground colors, F

Pixels with unknown B not included Non-linear least squares, using forward differencing

for Jacobian

views pixels

i j

jijijjiji BαFαCO1 1

2))(1()(),( xxFx

Page 13: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Additional Penalty Terms Favor control points at strong edges

define potential field around each edgel

Discourage large motions (>2 pixels) helps avoid degenerate curves

Page 14: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Naïve object insertion (no matting)

Page 15: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Object insertion with Boundary Matting

Page 16: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Naïve object insertion (no matting)

Object insertion with Boundary Matting

Page 17: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Naïve object insertion (no matting)

Object insertion with Boundary Matting

boundaries calculated with subpixel accuracy

Page 18: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Samsung commercial sequence

Page 19: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Naïve object insertion (no matting)

Object insertion with Boundary Matting

Page 20: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Boundary Matting Naïve method

Page 21: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Boundary Matting Naïve method

Page 22: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

boundary mattingboundary matting (sigma = 13)boundary matting (sigma = 26)compositebackgroundno matting

Synthetic Noise

Page 23: Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Interactive Visual Media Group Microsoft Research {sbkang,szeliski}@microsoft.com Dept. of Computer

Concluding Remarks

Boundary Matting better view synthesis refines stereo at occlusion boundaries subpixel boundary estimation

Future work incorporate color statistics extend to dynamic setting