software image stabilization comes of age

39
Software image stabilization comes of age Michal Šorel

Upload: ophira

Post on 22-Jan-2016

43 views

Category:

Documents


0 download

DESCRIPTION

Software image stabilization comes of age. Michal Šorel. Removing camera motion blur. Alternative to OIS (optical image stabilization) systems Camera motion, not subject motion. Talk outline. How to describe the blur? (velocity field, space-variant PSF ) Common setups Single blurred image - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Software image stabilization comes of age

Software image stabilization comes of age

Michal Šorel

Page 2: Software image stabilization comes of age

Removing camera motion blur

• Alternative to OIS (optical image stabilization) systems

• Camera motion, not subject motion

Page 3: Software image stabilization comes of age

Talk outline

• How to describe the blur? (velocity field, space-variant PSF )

• Common setups– Single blurred image– Multiple blurred images– Multiple underexposed/noisy images– One blurred and one underexposed image

• Perspectives

Page 4: Software image stabilization comes of age

Stabilizer of 3D camera rotation

• Rigid body – 6 degrees of freedom

• Natural coordinate system

• Camera rotation is mostly dominant

• Blur is independent of scene depth (that is why optical image stabilizers can work)

Page 5: Software image stabilization comes of age

Camera rotates downwards ↓

Page 6: Software image stabilization comes of age

Velocity field

Page 7: Software image stabilization comes of age

Rotation about optical axis

Page 8: Software image stabilization comes of age

General 3D rotation

Page 9: Software image stabilization comes of age

• PSF h ... depends on position (x,y)

• Generalized convolution

• Convolution case – h is called convolution kernel or convolution mask

Space-variant PSF

Page 10: Software image stabilization comes of age

PSF for camera shake

h(s,t; x2,y2)

h(s,t; x1,y1)

h(s,t; x3,y3)

(x1,y1)(x2,y2)

(x3,y3)

Page 11: Software image stabilization comes of age

Talk outline

• How to describe the blur? (velocity field, space-variant PSF )

• Common setups– Single blurred image– Multiple blurred images– Multiple underexposed/noisy images– One blurred and one underexposed image

• Perspectives

Page 12: Software image stabilization comes of age

Single image deblurring

• Rob Fergus building on the work of James Miskin

• Bayesian approach• Approximation – conditional distributions of

PSF and image are considered independent• Priors on image gradients and blur kernels as

a mixture of Gaussians and exponential functions

Page 13: Software image stabilization comes of age

Bayesian approach

Page 14: Software image stabilization comes of age

Image prior

Intensity histogram Gradient histogram

Page 15: Software image stabilization comes of age

Image prior

Gradient histogram

Page 16: Software image stabilization comes of age

Image priors

Tikhonov regularization

TV regularization

Page 17: Software image stabilization comes of age

PSF prior

Page 18: Software image stabilization comes of age

Functional to minimize

Page 19: Software image stabilization comes of age

Rob Fergus (Example I)

Page 20: Software image stabilization comes of age

Rob Fergus (Example II)

Page 21: Software image stabilization comes of age

Talk outline

• How to describe the blur? (velocity field, space-variant PSF )

• Common setups– Single blurred image– Multiple blurred images– Multiple underexposed/noisy images– One blurred and one underexposed image

• Perspectives

Page 22: Software image stabilization comes of age

Multiple blurred images

• Multichannel blind deconvolution

• Convolution model of blurring

• Solved by minimization of

Page 23: Software image stabilization comes of age

Multiple blurred images

Page 24: Software image stabilization comes of age

Talk outline

• How to describe the blur? (velocity field, space-variant PSF )

• Common setups– Single blurred image– Multiple blurred images– Multiple underexposed/noisy images– One blurred and one underexposed image

• Perspectives

Page 25: Software image stabilization comes of age

Multiple noisy images

• Noise variance of the sum of N images is the same as of the original image

• The difficult part is registration

• Main problem slow read-out

N imagestime t’=t/N

noise variance σ2/N

1 imagetime t =1s

noise variance σ2

Page 26: Software image stabilization comes of age

Talk outline

• How to describe the blur? (velocity field, space-variant PSF )

• Common setups– Single blurred image– Multiple blurred images– Multiple underexposed/noisy images– One blurred and one underexposed image

• Perspectives

Page 27: Software image stabilization comes of age

Blurred + underexposed image

• noisy ~ underexposed (exposure time changes contrast)

• patented in 2006

• since 2006 - several papers assuming convolution model

• our space-variant version sent to BMVC 2008

Page 28: Software image stabilization comes of age

Deblurring algorithm

Blurredimage

Noisyimage

Page 29: Software image stabilization comes of age

Image registration

• Small change of camera position – small stereo base

• Static parts of the scene can be modelled by projective tranform found by RANSAC

• Lens distortion can be neglected

• Less important parts of scene can move

Page 30: Software image stabilization comes of age

Blurred + underexposed results

Page 31: Software image stabilization comes of age

Blur kernel adjustment

• Regions lacking texture

• Regions of pixel saturation

Page 32: Software image stabilization comes of age

Restoration

• Minimization of functional

• PSF h interpolated from estimated convolution kernels

Page 33: Software image stabilization comes of age

Shopping center (details)

Page 34: Software image stabilization comes of age

Bookcase example

Page 35: Software image stabilization comes of age

Bookcase (details)

Page 36: Software image stabilization comes of age

Summary/Perspectives

• Denoising – artifacts or readout problems• Single image approach – takes time,

imprecise PSF, unable to distinguish intentional depth of focus, limited to convolution model

• Multiple blurred images – computationally expensive, fewer artifacts

• Blurred + underexposed image – fastest,algorithm for space-variant deblurring exists

Page 37: Software image stabilization comes of age

Discussion, questions...

Michal ŠorelInstitute of Information Theory and Automation of the ASCR

[email protected]

Page 38: Software image stabilization comes of age

Project ?

• Segmentation of moving objects from noisy-blurry image pair

• Restoration from one noisy and two blurred images

• Space-variant single image deblurring

• Combination with demosaicing to get full resolution image

• Implementation in fix-point arithmetic? ...

Page 39: Software image stabilization comes of age

Cause of blur

• Long exposure time -> apparent image motion more than about half pixel

• Why do we need long time?– enough photons to avoid quantization and shot

noise– Small aperture to achieve high depth of focus– Small aperture because of tele lens

construction limitations