Cast Shadow Segmentation using Invariant Color Features

1
Shadow Removal

• Cast Shadow Segmentation using Invariant Color
  Features

Elena Salvador, Andrea Cavallaro, Touradj
Ebrahimi March 2003
Daniel Spinosa Brown University 2004
2
The Problem
? Sometimes you dont want shadows

• Aesthetic
• Image Segmentation

?
3
Goal
? Segment Shadows In Images ?
4
Previous Approaches
Model-based
Property-based

• Requires a priori knowledge!
• Geometry . . .
• Scene
• Objects
• Lighting

• Utilizes
• Brightness
• Color of shadows
• Invariant images
• Geometry

Detection of Cast Shadows for Object
Segmentation Stauder et. al.
Good for specific applications
Security Camera Satellite / Aerial
? Requires regular background, video sequence
5
New Approach
C
B
Unlikely Shadow Points (different invariant
color)
Candidate Shadow Points (color edge map)
Original image
B - C More Likely Cast Shadow Points
Shadow Contours
Refinement by Geometric analysis
6
(real life) Results
7
Plan
Week1 Change projects and present again Week2
Repeat Mid-project
Presentation -Thanksgiving- Week3 Finish Core
of Algorithm Week4 Finish Reintegration and
begin unanticipated work Week5 Complete
unanticipated work tweak system
Final Project Presentation
8
Plan
Week1 Edge detection, invariant edge detection
? shadow hypothesis Week2 Refinement of shadow
contour image, Begin shadow contour
completion -Thanksgiving- Week3 Finish shadow
contour completion, shadow region
extraction Week4 System testing ? Tweak
algorithm / begin unanticipated work Week5
Finish system, Further testing
Final Project Presentation