Rapid Object Detection using a Boosted Cascade of Simple Features

Report
Rapid Object Detection using a
Boosted Cascade of Simple Features
Paul Viola, Michael Jones
Conference on Computer Vision and
Pattern Recognition 2001 ( CVPR 2001 )
Outline
•
•
•
•
•
Introduction
Features
Learning Classification Functions
The Attentional Cascade
Result
Introduction
Three Contribution
• New image representation
- Integral image
• Method for constructing a classifier
- Selecting a small number of important features
using AdaBoost
• Method for combining classifiers
- In a cascade structure
Features
Three Kind of Features
• Two-rectangle
• Three-rectangle
• Four-rectangle
• Feature value = sum of pixel value in white area
- sum of pixel value in black area
Integral Image
• Integral Image
Rectangular Sum
Rectangular
Sum Location
A
1
B
2-1
C
3-1
D
4+1-(2+3)
Learning Classification
Function
Learning Classification
Function
• Very small number of features can form an effective
classifier
• Select best classifier feature
• Weak classifier
AdaBoost algorithm
AdaBoost algorithm
Learning Result
• A frontal face classifier
- 200 features (among 180,000)
- Detection rate: 95%
- False positive rate: 1/14084
- 0.7s to scan an 384*288 pixel image
• First feature selected
- The eyes is often darker than the nose and cheeks
• Second feature selected
- The eyes are darker than the bridge of the nose
The Attentional Cascade
Cascade
Training a cascade of
classifiers
• Tradeoffs
o Features↑ ↔ detection rates ↑
o Features↑ ↔ computational time ↓
• Constructing stages
o Training classifiers using AdaBoost
o Adjust the threshold to minimize false negative
Result
Result
• Face training set
o
o
o
o
4916 faces image
24*24 pixels
9544 image
350 million sub-windows
• The complete face detection cascade has
o 38 stages
o 6061 features
o 15 times faster than current system
Performance
Performance
Result
Thank you for your attention!

similar documents