12/6/2013

3 Final Problems
Photography in low light Photography in bad weather Photo forgeries
G. Petschnigg, M. Agrawala, H. Hoppe, R. Szeliski, M. Cohen, K. Toyama

Digital Photography with Flash and No-Flash Image Pairs

E. Eisemann and F. Durand

Photography in Low Light

Using available ambient light:
+ natural lighting - high noise - color needs white balancing - blur
No-flash

Adding Lighting Shows Details

Using flash:
+ details + color + low noise - flat/artificial - flash shadows - red eye
Flash

12/6/2013

Flash + No-Flash
Combine best aspects of each image

Flash + No-Flash Approach

Use no-flash image to relight flash image (Petschnigg), or use flash image to relight no-flash image (Eisemann)

No-flash

+ original lighting + details/sharpness + noise removal + color

No-flash

Result Flash Result

Acquisition Process

Acquisition Process

Lock Focus & Aperture

time

Lock Focus & Aperture

No-Flash Image Large Sensor Gain/ISO

1/30 s ISO 3200 time

12/6/2013

Acquisition Process

Eisemann and Durand Algorithm

Use color from flash image after inverse white balancing

Lock Focus & Aperture

No-Flash Image Flash Image Large Sensor Gain/ISO Low Sensor Gain/ISO
1/30 s ISO 3200 1/125 s ISO 200 time

Flash shadow detection and deletion color after white balancing texture detail

Eisemann and Durand Algorithm

Recombination

*
Flash shadow detection and deletion

=
Fine-scale Flash Intensity Result

Coarse-scale No-flash

Recombination: Large scale * Detail = Intensity

12/6/2013

Recombination
shadow removal

Results

*
Intensity Result Color Flash

~ ~
Result

No-flash

Recombination: Intensity * Color = Original

Flash

Results

Poor Result

No-flash

No-flash

Flash

Result

Flash

12/6/2013

Photographs in Bad Weather

Haze

Mist

CVPR 2009

Fog

Rain
Images Courtesy : Steve and Carol Sheldon

Aerial Perspective
aka Atmospheric Perspective Objects farther in the distance appear less saturated (whiter) and less sharp (blurrier) than those nearby The more atmospheric particles between the viewer and a distant object, the more light is scattered
Low contrast

12/6/2013

Distant Objects are Desaturated

Aerial Perspective

Leonardo, Virgin and St. Anne, 1510

Color Perspective

Distant objects tend toward blue, near objects toward red

12/6/2013

Medium transmission

Heuristic: Haze-free images have higher contrast than hazy images

Most local patches in haze-free outdoor images that do not contain sky contain some pixels that have very low intensities in at least one color channel!

12/6/2013

Haze-free images have most pixels in dark channel near 0

12/6/2013

Daytime, outdoor landscapes or cityscapes from Flickr

12/6/2013

The intensity of the dark channel is an approximation of the thickness of the haze use it to estimate J, A, and t

10

12/6/2013

Results

input

Results

Results

recovered image

depth

11

12/6/2013

Results

input

recovered image

Digital Image Forensics: Detecting Forgeries

1. Detecting photorealistic graphics

2. Detecting manipulated images

12

12/6/2013

CG vs. Real: Can You Do It?

fakeorfoto.com

Digital Image Forensics: Detecting Forgeries http://www.life.com/archive/realfake

I got 5 out of 12 right

Chance = 6 out of 12

I got 6/10
Chance = 5/10

Detecting Forgery -- Why It Matters: Trust

Examples collected by Hany Farid: http://www.cs.dartmouth.edu/farid/research/tampering.html

Iconic Portrait of Lincoln (1860) Lincolns head on John Calhouns body

1989 composite of Oprah and Ann-Margret (without eithers permission)

13

12/6/2013

1930s: Stalin had disgraced comrades airbrushed out of his pictures

http://www.newseum.org/berlinwall/commissar_vanishes/index.htm

http://www.cs.dartmouth.edu/farid/research/digitaltampering/

2008

2000: black students face inserted into UW magazine

Pulitzer Prize winning photograph of Kent State killing (1970)

http://www.cs.dartmouth.edu/farid/research/digitaltampering/

14

12/6/2013

Published photo

Caption: Actress and Anti-war activist Jane Fonda speaks to a crowd of Vietnam veterans, as activist and former Vietnam vet John Kerry listens and prepares to speak next concerning the war in Vietnam. (AP Photo)

2003: Long-time staff photographer for LA Times was fired for this one

Kerry at Rally for Peace 1971

Fonda at rally in 1972

Detecting Forgery: Cloning

Exposing Digital Forgeries by Detecting Duplicated Image Regions
A.C. Popescu and H. Farid Technical Report, TR2004-515, Dartmouth College, Computer Science

Detecting Forgery: Retouching

Exposing Digital Forgeries in Color Filter Array Interpolated Images
A.C. Popescu and H. Farid IEEE Transactions on Signal Processing, 53(10):3948-3959, 2005
2005: Pres Bush scribbles a note to C. Rice during UN Security Council Meeting

15

12/6/2013

Demosaicing Prediction
In demosaicing, RGB values are filled in based on surrounding measured values Filled in values will be correlated in a particular way for each camera Local tampering will destroy these correlations

Detecting Forgery: Lighting/Shadows

Exposing Digital Forgeries by Detecting Inconsistencies in Lighting
M.K. Johnson and H. Farid ACM Multimedia and Security Workshop, New York, NY, 2005

Farid: Photo Fakery and Forensics 2009

Detecting Forgery: Lighting/Shadows

Exposing Digital Forgeries by Detecting Inconsistencies in Lighting
M.K. Johnson and H. Farid ACM Multimedia and Security Workshop, New York, NY, 2005

Estimating Lighting Direction

1 Method: 2D direction from occluding contour
Provide at least 3 points on occluding contour (surface has 0 angle in Z direction) Estimate light direction from brightness
Estimate

Ground Truth

16

12/6/2013

Detecting Inconsistencies in Lighting

Real photo

Detecting Forgery: Lighting/Shadows

Fake photo

Lighting: Specular Highlights in the Eye

Estimating Lighting from Eyes

M.K. Johnson and H. Farid, Exposing Digital Forgeries Through Specular Highlights on the Eye, 2007

17

12/6/2013

Summary
Digital forgeries are an increasingly major problem as it becomes easier to fake images A variety of automatic and semi-automatic methods are available for detection of well-done forgeries
Checking lighting consistency Checking demosaicking consistency (for high quality images) Checking JPEG compression level consistency (for low quality images)

18