Removing Weather Effects from - PowerPoint PPT Presentation

1 / 13
About This Presentation
Title:

Removing Weather Effects from

Description:

Removing Weather Effects from Monochrome Images Srinivasa Narasimhan and Shree Nayar Computer Science Department Columbia University – PowerPoint PPT presentation

Number of Views:78
Avg rating:3.0/5.0
Slides: 14
Provided by: Sriniv3
Category:

less

Transcript and Presenter's Notes

Title: Removing Weather Effects from


1
Removing Weather Effects from Monochrome
Images
Srinivasa Narasimhan and Shree
Nayar Computer Science Department Columbia
University IEEE CVPR Conference December 2001,
Hawaii, USA Sponsors DARPA HID, NSF
2
Contrast Degradation in Bad Weather
Rain
Fog
How does scene contrast degrade in bad
weather ? How can scene contrast be
restored from bad weather images ?
3
Image Processing Does Not Suffice
Histogram Equalized Images
Weather Effects are Depth Dependent
4
Prior Methods for Contrast Restoration
Weather Information
Clear-day Scene Intensity/Color
Method
Scene Depth
Predicted Weather PSF
Required
Computed
Yitzhaky, Kopeika 98
Gaussian distribution assumed
Oakley, Tan, Satherley 98,01
Required
Not Required
Wavelength Independent Scattering
Nayar, Narasimhan 99 Narasimhan, Nayar 00
Computed (Color Images Required)
Computed
Computed
Not Required
Computed
OUR GOAL
5
Scattering Models Attenuation and Airlight
Diffuse Skylight
Sunlight
Object
Observer
d
Diffuse Ground Light
Attenuation
Airlight
6
Contrast Degradation in Bad Weather
Irradiance Attenuation
Airlight


Scattering Coefficient
Reflectance
Horizon Brightness
Depth
(1)
(2)
Contrast between Iso-Depth points , P
and P
Contrast Decay Exponential in Scene Depth
7
Depth Edges vs. Reflectance Edges
Mild Fog
Denser Fog
Reflectance Edge
Depth Edge
Normalized SSD of Depth Edge Neighborhood
Normalized SSD of Reflectance Edge
Neighborhood
8
Edge Classification from Weather Changes
Edge Classification
Mild Fog
Denser Fog
Reflectance Edge Depth Edge
9
Scene Structure from Weather Changes
Irradiance under versus
Irradiance under Linear
All Scene points at Depth 1
All Scene points at Depth 2
Scaled Depth
10
Depth Map from Two Weather Conditions
Mild Fog, 5 PM (Input)
Computed Depth Map (Output)
  • Comparing with Prior Methods
  • Color Images Not Needed
  • Works for Wider Range of
  • Weather Conditions

Denser Fog , 5 30 PM (Input)
11
Weather Removal Using Scene Structure
Contrast Restored Image (Output)
Dense Fog, 530 PM (Input)
Computed Depth Map (Input)
Histogram Equalized Image (For comparison)
12
Different amounts of Fog removed from
different Depths
13
A De-Weathering System
System Initialization Computing Scene
Structure
Video Frame (Weather 1)
Detect Significant Weather Change
Scene Structure
Video Frame (Weather 2)
Continuous De-Weathering Using Scene Structure
Contrast Restored Video Frame
Scene Structure
Write a Comment
User Comments (0)
About PowerShow.com