Object detection

1
Object detection

• Presented by Minh Hoai Nguyen
• Date 28 March 2007

2
Object detection?
3
What we want
Miss a face!
4
Happy face!
5
Scanning window
Train a classifier on a fixed size window
6
Outline

• Object Detection Using the Statistics of Parts
• Schneiderman, H. Kanade, T. CVPR00, IJCCV04
• Robust Real-time Face Detection
• Viola, P. Jones, M. CVPR01, IJCV04

7
Bayes optimal classifier

• Image is defined by n attrs x1,x2,,xn

8
Naïve Bayes Assumption

• Assume x1,x2,,xn are cond. independent.

• Problem this might be a bad assumption

9
Independent groups/parts

• How to divide x1,x2,,xn into ind. groups?
• Image pixels are highly correlated.
• Represent image by Wavelets instead.

10
Wavelet transform
10 filter responses for each original pixel.
LH

• Wavelet transform is fully invertible.
• Partially de-correlate natural imagery
• More independence, easier to design parts

11
Designing parts

• Assumption
• Each wavelet coefficient only depends on few
  others.
• Group those coefficients into parts.
• Parts
• 17 types, manually defined.
• Each part contains 8 coefficients.

12
Categories of parts
Intra-subband
Local operator
Inter-frequency
Local operator
Parts
Inter-orientation
Local operator
Inter-frequency/ Inter-orientation
Local operator
Slide credit Nicholas Chan
13
Final form of detector
How to compute these statistics?
Count!
14
Multiple poses?

• Other tricks
• Not going to talk about.

15
Reported results for faces

• Kodak dataset
• Test set 17 images, 46 faces, 36 profile views.

16
A bigger dataset

• From multiple sources 208 images, 441 faces,
  about 347 profiles.

17
Robust Real-time Face Detectionby Viola,P.
Jones, M.
18
Cascade of classifiers

• Most places do not have faces!

19
Simple features
Box filters Approximation of Harr-wavelets
20
Learning the cascade

• AdaBoost
• Weak classifiers are box filters

21
Learning cascade stages

• Using AdaBoost to train each stage
• Adjust threshold to minimize false negatives.
• Adding features until target detection and false
  positive rates are met (determined by CV)

22
Learned cascade

• The whole cascade
• 38 stages
• 6000 features in total
• On dataset with 507 faces and 75 millions
  sub-windows, faces are detected using 10 feature
  evaluations on average.
• On average, 10 feature evals/sub-window

23
Reported ROC curve
24
Comparison results
25
The end