Paper #0394 - Single Image Dehazing

Single Image Dehazing

Raanan Fattal

In this project we present a new method for estimating the optical transmission in hazy scenes given a single input image. Based on
this estimation, the scattered light is eliminated to increase scene visibility and recover haze-free scene contrasts. In this new
approach we formulate a refined image formation model that accounts for surface shading in addition to the transmission function. This
allows us to resolve ambiguities in the data by searching for a solution in which the resulting shading and transmission functions
are locally statistically uncorrelated. A similar principle is used to estimate the color of the haze. Results demonstrate the new
method abilities to remove the haze layer as well as provide a reliable transmission estimate which can be used for additional
applications such as image refocusing and novel view synthesis.

A new work on image dehzaing using color-lines can be found here, and on estimating the atmospheric light vector here.

Results Page

2008 Siggraph paper (12.7MB PDF)

A talk at MSR

VIDEO: refocusing & novel views (17MB AVI)

A matlab demo code

Information about licensing this technlogy can be found here.

 


 

Red Bricks House

 

Input Image
Output Image
Depth
Modified Fog Layer
Refocused

Refocused

 


 

Wheat Cones

 

Input Image
Output Image
Depth Image
Raw Estimate (before extrap.)
Refocued
Shading Component l(x)
Transmission t(x)
Extrapolated Trans.
Certainty (1/Var(sigma_t))

 


Synthetic Example

 

Input Image
Transmission t(x)
Shading l(x)
Output
Transmission t(x)
Shading l(x)
Output

 


 

Aerial & Underwater Photography

 

Input
Output
The airlight-albedo mutual component, Eta (black pixels discarded)

 

Input

Output

 

 


 

Forest

 

Input Image
Output Image
Denser Fog
IA(x)
IR(x)
h(x)

 


People (watching a show)

 

Input Image
Output Image
Refocused


Mountain

 

Input Image
Output Image
Depth


 

Pumpkins

 

Input Image
Output Image
Refocued
Depth

 


 

Comparisons

 

Landscape

(Image taken from Schechner et al 2001)

 

Input Image
Polarization-Based, Schechner et al. 2001
Output (our)
Dark-Object Subtraction, Chavez 1988
Dark-Object Subtraction, Chavez 1988
Histogram Equalization
Multi-Scale Detail Enhancement, Fattal et al. 2007
Unsharp Mask
Gamma Correction

 

New York

(Image taken from Schechner et al 2001)

 

Input Image
Polarization-Based, Schechner et al. 2001
Output (our)
Dark-Object Subtraction, Chavez 1988
Unsharp Mask
Histogram Equalization

 

Additional Fields

Estimated Transmission t(x)
Extrapolated Trans.
Certainty (1/Var(sigma_t))
IA(x)
IR(x)
H(x)