study active refocusing of images and videos

35
Francesc Moreno-Noguer Computer Vision Lab. Ecole Polytechnique Fédérale de Lausanne Peter N. Belhumeur Shree K. Nayar Columbia University SIGGRAPH 2007 Active Refocusing of Images and Videos study

Upload: chiamin-hsu

Post on 18-May-2015

475 views

Category:

Documents


1 download

DESCRIPTION

study

TRANSCRIPT

Page 2: study Active Refocusing Of Images And Videos

Abstract

Use an active illumination method for depth estimation from a single image

Acquired ImageComputed Depth

NearFar

Refocused (Near)Refocused (Far)Alternate Lighting

Page 3: study Active Refocusing Of Images And Videos

Outlines

Introduction Related Work Overview Projection Dot Defocus Analysis Dot Removal & Depth Estimation Realistic Refocusing Result Limits and Conclusions

Page 4: study Active Refocusing Of Images And Videos

Introduction of Refocusing

Challenges of Active Refocusing Dynamic scenes

Depth Estimation be done in a single frame

Active illumination Full resolution depth map Projection Dot removal

Partial Occlusions

xfy k captured

blur kernels at depth k

In-focus

Page 5: study Active Refocusing Of Images And Videos

RELATED WORK

Page 6: study Active Refocusing Of Images And Videos

Relative Work: Depth EstimationPassive Methods

Active Illumination Methods

Shape from shading Cannot handle depth

discontinuities

Coded Aperture [Levin et al. SIGGRAPH 07] Cam. H.W. modify Require Light Source

Pattern

Structured Light [Salvi et al. Pattern Recognition ,04] No pattern removal

Projector Temporal Defocus [Zhang & Nayar SIGGRAPH06]

Page 7: study Active Refocusing Of Images And Videos

Relative Work: Digital RefocusingRefocusing Given Depth

Light Field Photography

Synthesis Images: Ray Tracing [Cook SIGGRAPH84] Require complete 3D model

Real Images: Convolution [Photoshop; IrisFilter] Partial Occlusions Problem

Light Field Camera [Ng SIGGRAPH05] Cam. H.W. modify Resolution losses

Dappled Photography [Veeraraghavan SIGGRAPH07] Cam. H.W. modify Layer

Page 8: study Active Refocusing Of Images And Videos

Sparse Depth Map

Acquired Image

Dots Removed

Color Segmentation

Merged Segmentation

Dense Depth

Dots Depth Estimation by

Calibration

Dots Removal

Matting

Depth Map Completion using Segmentation

Depth Estimation

Page 9: study Active Refocusing Of Images And Videos

Depth Map

Realistic Refocusing

Dots Removed

Focal plane,Apertures,Window size of dots

Page 10: study Active Refocusing Of Images And Videos

PROJECTION DOT DEFOCUS ANALYSIS

Page 11: study Active Refocusing Of Images And Videos

System Design

Camera & Projector Coaxial have same Optical Axis

Projector

Page 12: study Active Refocusing Of Images And Videos

Blur Circle Diameter, D

fc uurfD

112

uf

Dr

u

fcv

v

fwDD c

w

w

with dot size w (in the projector plane)

Page 13: study Active Refocusing Of Images And Videos

Blur Circle Radiance, I

21I

f

p

uu

uf

Dr

u

fcv

2

1

2

I

vr

wu

u

uw

f

w

w

with dot size w (in the projector plane)

based on Image Irradiance Equation derived in [Horn 86]

Page 14: study Active Refocusing Of Images And Videos

Camera images of dot of 3*3 pixels projected onto different depths

Page 15: study Active Refocusing Of Images And Videos

Camera images of dot of 3*3 pixels projected onto different depths

Page 16: study Active Refocusing Of Images And Videos

DOT REMOVAL AND DEPTH ESTIMATION

Page 17: study Active Refocusing Of Images And Videos

Calibration Patches

Estimated

Sparse Depth MapDepth 1Depth 2 …

X =

Page 18: study Active Refocusing Of Images And Videos

Calibration Patches

Estimated

Sparse Depth MapDepth 1Depth 2 …

Page 19: study Active Refocusing Of Images And Videos

Calibration Patches

Estimated

Sparse Depth MapDepth 1Depth 2 …

Page 20: study Active Refocusing Of Images And Videos

Depth Estimation - ux

Non-textured Surface

Textured Surfaces (texture by itself introduces brightness variation)

riN

j ijcic ,, II

riN

jijc

iix uu ,Ivarminarg|

ici

ix uu ,Iminarg|

based on Unsupervised Learning Alg. [Figueiredo and Jain IEEE02]

Page 21: study Active Refocusing Of Images And Videos

DEPTH MAP COMPLETION USING SEGMENTATION

Page 22: study Active Refocusing Of Images And Videos

Depth Map Completion

Sparse Depth Map

Over-Segmentation

Mean-Shift[Comaniciu & Meer 02]

Iterative Merging

Page 23: study Active Refocusing Of Images And Videos

Depth Map Completion – Iterative Merging

Loop: Apply Greedy alg. to group segments Merge the two most

similar neighboring segments

Re-computes the features of the new merged segment

Iterative Merging

Page 24: study Active Refocusing Of Images And Videos

Similarity between Segments

Sim(i,j)=λC∙dist(Ci,Cj)+λD∙dist(Di,Dj)+λT∙dist(Ti,Tj)

Color C Depth D Texture T

Page 25: study Active Refocusing Of Images And Videos

Depth Map Completion – Refine the Depth Disc.

Noisy Depth Map

Matting Algorithm[Wang & Cohen

05]

Page 26: study Active Refocusing Of Images And Videos

REALISTIC REFOCUSING

Page 27: study Active Refocusing Of Images And Videos

Challenge of Refocusing Partial occlusions

Different parts of the lens may see different views at an object boundary

Create missing region by detecting discontinuities in depth map and extending the occluded surface using texture synthesis

Foreground/background transitions Pixels at depth discontinuities may

receive contributions from the fr. and bg. Blend fr./bg. images within the

boundary region

A)(RRR 1A GCFC

)A-(1RARR GCFC

Page 28: study Active Refocusing Of Images And Videos

Realistic Refocusing produces better results than existing approaches

Original

Realistic Refocusin

g

Canon + wide

aperture

Photoshop - blur

IrisFilter

Page 29: study Active Refocusing Of Images And Videos

Partial Occlusions

Page 30: study Active Refocusing Of Images And Videos

Refocusing with Alpha Maps

R RC Є F* += *R

C Є B

Background (B)

Boundary (C)

Foreground (F)

RC Є F

RC Є B

Page 31: study Active Refocusing Of Images And Videos

RESULT

Page 32: study Active Refocusing Of Images And Videos

Limitations

Due to Active Illumination

Translucent objects exhibit subsurface scattering

Blurred dots are too weak to detect Very dark Highly inclined surface

(> 70°)

Poor in outdoor with strong sunlight

ex: the ball and the table are assigned diff. depths due to errors on segmentation errors

Page 33: study Active Refocusing Of Images And Videos

Limitations

Due to sparse dots Sparsity of the depth estimation

Errors in the initial segmentation of the image

ex: incorrect depth due to segmentation err.

Page 34: study Active Refocusing Of Images And Videos

Conclusions

Contribution Future Work

An active illumination depth estimation with single Single Frame, Complete

Depth Map, Texture/Textureless scenes

Projected Light Patterns are Removed

High resolution refocusing of images and videos

Incorporate the method into digital cameras

Use intra-red source for projecting the dot patter to make the depth estimation more robust in the case of highly textured scenes

Page 35: study Active Refocusing Of Images And Videos

END