Main goal Face categorization problem
Object class recognition techniques have seen progress in recent years (ex: bag-of-words models)
Different people are considered different object classes Integration of SIFT features
Does SIFT help face recognition? To what degree? Target on a few people Applications
Face annotation in family albums Name-based photo search
Method Features: SIFT Bag-of-words representation for faces
Each SIFT feature is considered a “codeword” Build a dictionary based on training samples Each face is represented as a histogram over
codewords Learning: Naïve Bayes Classifiers
Vector quantization
…
…..
fre
que
ncy
codewords
SIFT feature vectors from training samples
128-d feature space
codeword 1codeword 2
codeword 3
Datasets The BioID Face Database (simple)
1521 images with 23 people Variety of illumination, background and face size
# of images per person
121 90 150 40
2 51 94 51
78 61 40 50
88 25 6 71
98 106 46 59
60 99 35 100
22 categories, 1613 images70% for training, 30% for testing
Datasets Faces in the Wild (challenging)
851 images, 10 people + 1 non-faces Extracted from news videos
http://www.cs.berkeley.edu/~millert/faces/faceDict/NIPSdict/
# of images per person
Agassi 74 Ryder 46
Arroyo 41 Schwarzenegger 72
Bush 82 Serena Williams 70
Gates 34 Venus Williams 39
Powell 83 non-faces 137
Putin 48
11 categories, 851 images70% for training, 30% for testing
81.82% 0.00% 0.00% 0.00% 0.00% 4.55% 4.55% 0.00% 4.55% 4.55% 0.00% 0.00% 83.33% 0.00% 0.00% 0.00% 0.00% 8.33% 8.33% 0.00% 0.00% 0.00% 4.00% 4.00% 68.00% 0.00% 4.00% 16.00% 4.00% 0.00% 0.00% 0.00% 0.00% 20.00% 0.00% 0.00% 80.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 8.00% 0.00% 0.00% 8.00% 68.00% 12.00% 0.00% 4.00% 0.00% 0.00% 0.00% 0.00% 0.00% 7.14% 0.00% 0.00% 85.71% 0.00% 7.14% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 100.00% 0.00% 0.00% 0.00% 0.00% 0.00% 9.09% 9.09% 0.00% 0.00% 4.55% 0.00% 77.27% 0.00% 0.00% 0.00% 0.00% 4.76% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 85.71% 9.52% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 8.33% 8.33% 83.33% 0.00% 0.00% 2.44% 2.44% 2.44% 2.44% 0.00% 0.00% 0.00% 2.44% 0.00% 87.80%
Average: 81.91%