lecture 33: computational photography

41
Lecture 33: Computational photography CS4670: Computer Vision Noah Snavely

Upload: tallis

Post on 16-Jan-2016

42 views

Category:

Documents


0 download

DESCRIPTION

CS4670: Computer Vision. Noah Snavely. Lecture 33: Computational photography. Photometric stereo. Limitations. Big problems doesn’t work for shiny things, semi-translucent things shadows, inter-reflections Smaller problems camera and lights have to be distant calibration requirements - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Lecture 33: Computational photography

Lecture 33: Computational photography

CS4670: Computer VisionNoah Snavely

Page 2: Lecture 33: Computational photography

Photometric stereo

Page 3: Lecture 33: Computational photography

LimitationsBig problems

• doesn’t work for shiny things, semi-translucent things• shadows, inter-reflections

Smaller problems• camera and lights have to be distant• calibration requirements

– measure light source directions, intensities

– camera response function

Newer work addresses some of these issues

Some pointers for further reading:• Zickler, Belhumeur, and Kriegman, "Helmholtz Stereopsis: Exploiting

Reciprocity for Surface Reconstruction." IJCV, Vol. 49 No. 2/3, pp 215-227. • Hertzmann & Seitz, “Example-Based Photometric Stereo: Shape

Reconstruction with General, Varying BRDFs.” IEEE Trans. PAMI 2005

Page 4: Lecture 33: Computational photography

Finding the direction of the light source

P. Nillius and J.-O. Eklundh, “Automatic estimation of the projected light source direction,” CVPR 2001

Page 5: Lecture 33: Computational photography

Application: Detecting composite photos

Fake photo

Real photo

Which is the real photo?

Page 6: Lecture 33: Computational photography

The ultimate cameraWhat does it do?

Page 7: Lecture 33: Computational photography

The ultimate cameraInfinite resolution

Infinite zoom control

Desired object(s) are in focus

No noise

No motion blur

Infinite dynamic range (can see dark and bright things)

...

Page 8: Lecture 33: Computational photography

Creating the ultimate cameraThe “analog” camera has changed very little in >100 yrs

• we’re unlikely to get there following this path

More promising is to combine “analog” optics with computational techniques• “Computational cameras” or “Computational photography”

This lecture will survey techniques for producing higher quality images by combining optics and computation

Common themes:• take multiple photos• modify the camera

Page 9: Lecture 33: Computational photography

Noise reductionTake several images and

average them

Why does this work?

Basic statistics: • variance of the mean

decreases with n:

Page 10: Lecture 33: Computational photography

Field of viewWe can artificially increase the field of view by

compositing several photos together (project 2).

Page 11: Lecture 33: Computational photography

Improving resolution: Gigapixel images

A few other notable examples:• Obama inauguration (gigapan.org)• HDView (Microsoft Research)

Max Lyons, 2003fused 196 telephoto shots

Page 12: Lecture 33: Computational photography

Improving resolution: super resolutionWhat if you don’t have a zoom lens?

Page 13: Lecture 33: Computational photography

D

For a given band-limited image, the Nyquist sampling theorem states that if a uniform sampling is fine enough (D), perfect reconstruction is possible.

D

Intuition (slides from Yossi Rubner & Miki Elad)

13

Page 14: Lecture 33: Computational photography

Due to our limited camera resolution, we sample using an insufficient 2D grid

2D

2D

14

Intuition (slides from Yossi Rubner & Miki Elad)

Page 15: Lecture 33: Computational photography

However, if we take a second picture, shifting the camera ‘slightly to the right’ we obtain:

2D

2D

15

Intuition (slides from Yossi Rubner & Miki Elad)

Page 16: Lecture 33: Computational photography

Similarly, by shifting down we get a third image:

2D

2D

16

Intuition (slides from Yossi Rubner & Miki Elad)

Page 17: Lecture 33: Computational photography

And finally, by shifting down and to the right we get the fourth image:

2D

2D

17

Intuition (slides from Yossi Rubner & Miki Elad)

Page 18: Lecture 33: Computational photography

By combining all four images the desired resolution is obtained, and thus perfect reconstruction is guaranteed.

Intuition

18

Page 19: Lecture 33: Computational photography

19

3:1 scale-up in each axis using 9 images, with pure global translation between them

Example

Page 20: Lecture 33: Computational photography

Dynamic RangeTypical cameras have limited dynamic range

Page 21: Lecture 33: Computational photography

HDR images — merge multiple inputs

Scene Radiance

Pixel count

Page 22: Lecture 33: Computational photography

HDR images — merged

Radiance

Pixel count

Page 23: Lecture 33: Computational photography

Camera is not a photometer!Limited dynamic range

• 8 bits captures only 2 orders of magnitude of light intensity• We can see ~10 orders of magnitude of light intensity

Unknown, nonlinear response • pixel intensity amount of light (# photons, or “radiance”)

Solution:• Recover response curve from multiple exposures, then

reconstruct the radiance map

Page 24: Lecture 33: Computational photography

Camera response function

Page 25: Lecture 33: Computational photography

Capture and composite several photosWorks for

• field of view• resolution• signal to noise• dynamic range• Focus

But sometimes you can do better by modifying the camera…

Page 26: Lecture 33: Computational photography

FocusSuppose we want to produce images where the desired

object is guaranteed to be in focus?

Or suppose we want everything to be in focus?

Page 27: Lecture 33: Computational photography

Light field camera [Ng et al., 2005]

http://www.refocusimaging.com/gallery/

Page 28: Lecture 33: Computational photography

Conventional vs. light field camera

Conventional camera

Light field camera

Page 29: Lecture 33: Computational photography

Light field camera

Rays are reorganized into many smaller images corresponding to subapertures of the main lens

Page 30: Lecture 33: Computational photography

Prototype camera

4000 × 4000 pixels ÷ 292 × 292 lenses = 14 × 14 pixels per lens

Contax medium format camera Kodak 16-megapixel sensor

Adaptive Optics microlens array 125μ square-sided microlenses

Page 31: Lecture 33: Computational photography
Page 32: Lecture 33: Computational photography

What can we do with the captured rays?Change viewpoint

Page 33: Lecture 33: Computational photography
Page 34: Lecture 33: Computational photography

Example of digital refocusing

Page 35: Lecture 33: Computational photography

All-in-focus imagesCombines sharpest parts of all of the individual

refocused imagesUsing single pixel from each subimage

Page 36: Lecture 33: Computational photography

All-in-focusIf you only want to produce an all-focus image, there

are simpler alternatives

E.g.,• Wavefront coding [Dowsky 1995]• Coded aperture [Levin SIGGRAPH 2007], [Raskar

SIGGRAPH 2007]– can also produce change in focus (ala Ng’s light field camera)

Page 37: Lecture 33: Computational photography

Why are images blurry?

Camera focused at wrong distanceDepth of field

Motion blur

How can we remove the blur?

Page 38: Lecture 33: Computational photography

Motion blurEspecially difficult to remove, because the blur kernel is

unknown

=

both unknown

Page 39: Lecture 33: Computational photography

Multiple possible solutions

=

Blurry image

Sharp image Blur kernel

=

=

Slide courtesy Rob Fergus

Page 40: Lecture 33: Computational photography

Priors can help

Priors on natural images

Image A is more “natural” than image B

Page 41: Lecture 33: Computational photography

Natural image statistics

Histogram of image gradients

Characteristic distribution with heavy tails