EE 368

1
Face Detection Project

• Nitin Kartiknitin.kartik_at_motorola.com
• EE368 Group 24
• Spring 2002,Stanford University
• Prof. Bernd Girod

2
Algorithm Design
Seconds
15

• Sub-Sample
• Color Rejection
• Black-and-White
• Correlation Coefficient
• Centroids

2
2
52
1
3
Step 1 Sub-sample

• 10x10 Reduction
• Averaging over pixels(Uniform Filter)
• Reduce Processing

4
Step 2 Color Rejection

• Trained from Examples
• Color Ratios
• Min/Max Table

Ratio Min Max Red Green 1.000 4.500 Green
Blue 0.843 12.000 Blue Red 0.0185 1.000
5
Step 3 Black-and-White

• No need for Color any more
• Average Red, Green, Blue Components
• Further Reduce Processing

6
Step 4 Correlation Coefficient

• Face Template (Averages)
• Compute Pixel-wise Correlation Coefficient
• Some Falsely Detected Faces

7
Step 4 Correlation Coefficient(False Alarms)

• Practical Classifiers

8
Step 5 Determine Centroids

• Hunting Algorithm
• Min / MaxWidth / Height
• Centroid (midX, midY)

9
Discarded Alternatives

• Sub-sampling with other factors
• Erosion / Dilation
• RMS-Based Rejection

10
Face Detection Project

• Nitin Kartiknitin.kartik_at_motorola.com
• EE368 Group 24
• Spring 2002,Stanford University
• Prof. Bernd Girod