Face Detection EE 368 Final Project

1
Face DetectionEE 368 Final Project
Group 11

• Donghyun Kim
• Jin Hyung Lee
• Joong Ho Won

2
Method

• Step 1 Skin Color Segmentation
• Step 2 Post Processing
• histogram intersection
• estimation of face geometry
• Step 3 Maximal Rejection Classification
• Step 4 Intersection of Step2 Step3
• Step 5 Refinement
• filtering to avoid false detection
• erase neighboring points to avoid multiple
  detection

3
Original Image
4
Skin Color Segmentation

• Skin Tone Color
• Bayesian Decision Rule

5
Post Processing
Enhancing Segmentation
Combining Histogram Intersection and Geometric
Estimation
6
Maximal Rejection Classification
of face images 1,620 of non-face images
1,849,953
7
Refinement
8
Result

• Setup Time
• Skin Probability Modeling 5 min
• Data Collection 30 min
• Training for MRC 10 min
• Detection Time 1.5 min
• for image size 1280960

9
Conclusion

• Face Detection accurate fast
• Existing Methods dramatic difference in speed
• Color Based Method Appearance Based Method
  accurate fast detection