2076

Single Image Dehazing Method

Fattal Raanan, HUJI, School of Computer Science and Engineering, Computer Science

tamir_huberman_raanan_fattal2.jpg

Eliminating scattered light to enhance images from a single input 

Categories

Imaging / Computer Graphics, Image Enhancement

Development Stage

Research on this project completed

Patent Status

Granted US Patent: 8,350,933  

 

 

 

Highlights

  • Improved method for recovering haze-free contrasts in images from a single input
  • Improves images from long-distance photography or in foggy conditions
  • Improved results compared with existing techniques, such as polarization-based dehazing
  • No need for multiple images or input of weather conditions

Our Innovation

  • New method for estimating the optical transmission in hazy scenes with minimal input requirements, a single image. Haze-free contrasts are recovered by using the optical transmission estimate to eliminate scattered light. The color of the haze is automatically estimated. The reliable transmission estimation produced enables extraction of scene dept.

Key Features

  • Can cope with images containing multiple layers of haze depth
  • Applications in aerial and underwater photography
  • May use the predicted transmission function to simulate a denser medium, e.g. sandstorm
  • May use results to produce novel view synthesis and image refocusing

Development Milestones

  • Future work includes incorporation of a deblurring mechanism regulated by the estimated transmission
  • Using blurriness data as additional information for estimating transmission function

The Opportunity

  • Haze-free long-distance or satellite images enhance cartography, and web mapping, land-use planning, archaeology, and environmental studies
  • Military and security applications
  • Improved consumer photography
For additional information: http://www.cs.huji.ac.il/~raananf/projects/defog/index.html

Patent Status

Granted US 8,350,933

Contact for more information:

Aviv Shoher
SVP BUSINESS DEVELOPMENT
+972-2-6586635
Contact ME:
Image CAPTCHA