Camouflage Breaking A Review of Contemporary Techniques

1
Camouflage BreakingA Review of Contemporary
Techniques

• Amy Whicker
• CSCE 867 Final Project

2

• What is camouflage?
• The process of masking the foreground to appear
  as though it is background.
• Camouflage related work can be divided into two
  areas
• Camouflage assessment and design
• Camouflage breaking
• Little has been researched in this area

3

• Why is camouflage breaking important?
• Military tactics
• Background subtraction
• Helps in the understanding of extraction of
  non-camouflaged objects
• Helps in developing algorithm to locates object
  in the foreground

4

• Camouflage Breaking Methods
• Multiple Camouflage Breaking by Co-occurrence
  and Canny
• Method developed by P. Nagabhushan and
  Nagappa U. Bhajantri
• Convexity-based Camouflage Breaking
• Method developed by Ariel Tankus and Yehezkel
  Yeshurun

5
Co-occurrence and Canny Method

• Part 1 Determine if there is a camouflaged
  object in the image.
• Create a gray level co-occurrence probability
  matrix.
• Assess the co-occurrence matrixs texture
  parameters.
• Part 2 Achieve effective visualization of
  camouflage objects.
• Repeatedly apply the Canny edge detection
  operator

6
Calculating the co-occurrence matrix

• Example from P. Nagabhushan and Nagappa U.
  Bhajantri. Multiple Camouflage Breaking by
  Co-occurrence and Canny.

7
Results of the Co-occurrence and Canny Method

• Images from P. Nagabhushan and Nagappa U.
  Bhajantri. Multiple Camouflage Breaking by
  Co-occurrence and Canny.

8
Convexity-based Method

• This method uses an operator (Darg) to create an
  output image whose intensity level is a
  reflection of the convexity of the original
  image.
• The Darg operator is defined by the sum of Yarg,
  rotated 0, 90, 180, and 270.
• Yarg is the y-derivative of the polar coordinates
  of the gradient argument of the original image.
  Yarg detects the zero-crossing of the gradient
  argument.

9
Convexity-based Method

• Images from Ariel Tankus and Yehezkel Yeshurun. A
  model for visual camouflage breaking.

10
Why Convexity?
Thayers principle of counter shading

1. A cylinder of constant albedo under top lighting.
   (b) A counter shaded cylinder under ambient
   lighting. (c) Thayers principle the combined
   effect of counter-shading albedo and top lighting
   breaks up the shadow effect (or convex intensity
   function).

• Images from Ariel Tankus and Yehezkel Yeshurun.
  Convexity-based Camouflage Breaking.

11
Convexity-based Method

• Though edge based methods have their advantages,
  this method overcomes some of the flaws of an
  edge-based approach such as,
• Sensitivity to illumination
• Scale
• Strong effect of the surroundings
• Cluttered or textured images

12
How does the Convexity-based method handle
changes in Illumination, Scale or Orientation?

• Images from Ariel Tankus and Yehezkel Yeshurun.
  Convexity-based visual Camouflage Breaking.

13
Convexity-based Method
Invariance to derivable strongly monotonically
increasing transformation of the gray-level
function.
Images from Ariel Tankus and Yehezkel Yeshurun.
Detection of regions of interest and camouflage
breaking by direct convexity estimation.
14
Convexity-based Method
Images from Ariel Tankus and Yehezkel Yeshurun. A
Model for Visual Camouflage Breaking.
15
Convexity-based Method
Images from Ariel Tankus and Yehezkel Yeshurun. A
Model for Visual Camouflage Breaking.
16
Convexity-based Method
Images from Ariel Tankus and Yehezkel Yeshurun.
Detection of regions of interest and camouflage
breaking by direct convexity estimation.
17
Convexity-based Method
Images from Ariel Tankus and Yehezkel Yeshurun.
Detection of regions of interest and camouflage
breaking by direct convexity estimation.
18
Convexity-based Method
Images from Ariel Tankus and Yehezkel Yeshurun.
Detection of regions of interest and camouflage
breaking by direct convexity estimation.
19
Conclusion

• Co-occurrence and Canny Method
• Advantages
• Simple
• Creates a good outline of the object
• Disadvantage
• Does not extract the object
• Must have the known background
• Only tested on synthetic images and may not be
  effective in real application

20
Conclusion

• Convexity-based Method
• Advantages
• Robust algorithm
• Precise in finding foreground objects
• Disadvantage
• Does not extract the object
• Threshold must be determined, which can change
  the results

21
References

• 1 P. Nagabhushan and Nagappa U. Bhajantri.
  Multiple Camouflage Breaking by Co-occurrence and
  Canny, University of Mysore, Manasa Ganotri,
  2004.
• 2 Ariel Tankus, Yehezel Yeshurun, and N.
  Intrator. Face Detection by Direct Convexity
  Estimation, Pattern Recognition Letters 18(9)
  (1997), 913-922.
• 3 Ariel Tankus and Yehezkel Yeshurun. Detection
  of regions of interest and camouflage breaking by
  direct convexity estimation, IEEE International
  Workshop on Visual Surveillance, pages 42-48,
  Bombay, India, January 1998. In conjunction with
  ICCV 1998.
• 4 Ariel Tankus and Yehezkel Yeshurun. A model
  for visual camouflage breaking, 1st IEEE
  International Workshop on Biologically Motivated
  Computer Vision (BMCV), pages 139-149, Seoul,
  Korea, May 2000.
• 5 Ariel Tankus and Yehezkel Yeshurun.
  Convexity-based camouflage breaking,
  International Conference on Pattern Recognition
  (ICPR), pages 454-457, Barcelona, Spain,
  September 2000.
• 6 Ariel Tankus and Yehezkel Yeshurun. Convexity
  Based Visual Camouflage Breaking, Computer Vision
  and Image understanding 82, (2001) 208-237.