Face Detection EE368 Final Project

1
Face Detection- EE368 Final Project -

• Stanford University, Dept. of EE
• Taesang Yoo, Youngjae Kim
• (Group 7)

2
Algorithm Outline
960x1280x3
240x320x3
240x320x1
Detect Skin-colors
Image Segmentation
Resize
Decision based on ratios, sizes
Yes
Face!!
No
Edge-Based Preprocessing
Eigenface method
Face!!
3
Skin colors?
- Fortunately, skin-colors form a cluster in
YCbCr color space - Thus, we can approximate the
skin color region with several lines.
4
Image segmentation
Segmentation based on connected areas
? many face candidate rectangles
?put strict restrictions on them ? face!!
5
Eigen face method (1)
- Applied to suspicious rectangles e.g. a
cluster of faces or non-face rectangles
Face samples (16x16 pixels) ?12 eigen faces
Suspicious rectangles
6
Eigen face method (2)
- Detectable range 21x21 80x80 pixels2 -
Resizing step 0.02 (5.00, 4.98, 4.96 1.30)
Detection example
Original
Reconstructed
7
Miscellany
- H-value of skin color is less than 0.1 or
greater than 0.9 ? (Trials and errors, but
useful to remove the background) - Edge based
preprocessing speeds up the process by enabling
the eigenface method to skip blank regions
8
Detection result
Training_1.jpg
9
Conclusions

• - A combination of skin color method eigenface
  method
• Reduced the false alarm rate using two filters
• Out of 163 faces in the 7 training images, 90
  with false alarm rate 5 was detected
• - Dependent on the training set by assumptions
  made from it (e.g. image size, of faces, colors)

10
References

• - Skin color detection
• Face Detection in Color Images using Wavelet
  Packet Analysis, C. Garcia, G. Zikos, G.
  Tziritas, Institute of Computer Science
  Foundation for Research and Technology, Greece.
• - Eigen face method
• Face Recognition Using Eigenfaces, Matthew A.
  Turk and Alex P. Pentland, Vision and Modeling
  Group, The Media Lab, MIT
• EE368 Lecture notes, Bernd Girod, Stanford
  University