Jan Sedmidubsky
April 7, 2014
Face Recognition Technology
Face Recognition Technology
Faculty of Informatics
Masaryk University
Brno, Czech Republic
fi-logo
1/11

Jan Sedmidubsky
•Face recognition technology
–
April 7, 2014
Face Recognition Technology
Motivation
00816_960530_rb00.jpg 00825_940307_fa00.jpg 00825_940307_fb00.jpg 00825_940307_hl00.jpg
00825_940307_hr00.jpg 00825_940307_pl00.jpg 00846_940307_hr_a00.jpg 00717_960530_fb00.jpg
query
face
Face
recognition
technology
ranked list of the most similar faces
“Jack Daniels” – query person name
Database of faces
00816_960530_rb00.jpg 00825_940307_fa00.jpg 00825_940307_hl00.jpg 00825_940307_hr00.jpg
00825_940307_pl00.jpg 00846_940307_hr_a00.jpg 00717_960530_fb00.jpg
2/11

Jan Sedmidubsky
•Components of face recognition technology
–Detection – automatic localization of faces in images
–Recognition – calculation of similarity of two faces
–Retrieval – search for the most similar faces from a large database
April 7, 2014
Face Recognition Technology
Motivation
3/11

Jan Sedmidubsky
•Our objective – to effectively and efficiently retrieve the most similar faces to a query face
•
•Approach
–Multi-technology – more effective detection/recognition
•Combination of existing techniques for detection and recognition
–Multi-query – more effective retrieval
•Query composed of a number of reference objects
–Scalability – more efficient retrieval
•Indexing MPEG-7 descriptors + effective re-ranking
April 7, 2014
Face Recognition Technology
Motivation
4/11

Jan Sedmidubsky
•Efficiency
–Time necessary for detection of faces in an image
–Time necessary for retrieval of the most similar faces
•Effectiveness
–Detection – recall 75%, precision 100%
–
–
–Recognition/retrieval – recall 50%, precision 40% when 8 relevant faces are in the database
–
April 7, 2014
Face Recognition Technology
Motivation
00788_941205_qr00.jpg 00816_960530_rb00.jpg 00825_940307_fa00.jpg 00825_940307_fb00.jpg
00825_940307_hl00.jpg 00825_940307_hr00.jpg 00825_940307_pl00.jpg 00846_940307_hr_a00.jpg
00717_960530_fb00.jpg
query
00788_941205_qr00.jpg 00816_960530_rb00.jpg
5/11

Jan Sedmidubsky
April 7, 2014
Face Recognition Technology
Multi-technology – Face Detection
•Multi-technology approach for face detection
–Combination of OpenCV, Luxand, Neurotechnology
–Agreement of at least 2 techniques out of 3
–
Low-quality dataset
High-quality dataset
Recall
Precision
Recall
Precision
OpenCV
55
89
92
86
Luxand
63
83
95
94
Neurotechnology
73
84
100
96
Combination
62
98
97
100
6/11

Jan Sedmidubsky
April 7, 2014
Face Recognition Technology
Multi-technology – Face Recognition
•Multi-technology approach for face recognition
–Combination of MPEG-7, Luxand, Neurotechnology
–Combination based on normalization of techniques
Low-quality dataset
(1k database faces)
High-quality dataset
(10k database faces)
Recall
precision=85%
Recall
precision=95%
Recall
precision=85%
Recall
precision=95%
MPEG-7
24
14
8
3
Luxand
23
16
14
0
Neurotechnology
12
11
53
51
Combination
31
24
54
51
7/11

Jan Sedmidubsky
April 7, 2014
Face Recognition Technology
Multi-query
•Multi-query
–Query composed of a number of reference objects
–
–
–
–
–Relevance feedback on 1.3M dataset:
•Manual selection of positive (correct) retrieval results
•Iterative search where positive results represent query objects
•1st iteration: R=P=6%, 5th iteration: R=P=30% (k=60)
00788_941205_qr00.jpg 00816_960530_rb00.jpg 00825_940307_fa00.jpg 00825_940307_fb00.jpg
00825_940307_hl00.jpg 00825_940307_pl00.jpg 00846_940307_hr_a00.jpg 00717_960530_fb00.jpg
query
00788_941205_qr00.jpg 00816_960530_rb00.jpg 00788_941205_qr00.jpg 00825_940307_fa00.jpg
00825_940307_fb00.jpg 00825_940307_hl00.jpg 00825_940307_hr00.jpg 00825_940307_pl00.jpg
00846_940307_hr_a00.jpg 00717_960530_fb00.jpg
query
00825_940307_hl00.jpg 00825_940307_pl00.jpg
8/11
00825_940307_hl00.jpg 00825_940307_hr00.jpg 00825_940307_hl00.jpg 00846_940307_hr_a00.jpg

Jan Sedmidubsky
April 7, 2014
Face Recognition Technology
Scalability
•Scalability
–Efficient search + re-ranking
–Candidate set retrieved by the metric MPEG-7 function
–Re-ranking of candidate set by multi-technology approach
9/11

Jan Sedmidubsky
April 7, 2014
Face Recognition Technology
Face Recognition Technology – API
•Technology can be controlled by API
–Management of faces/images:
•detectFaces, getImageFaces
•insertImage, insertFace, getAllFaces
–Retrieval:
•searchByFaceId, searchByFaceDescriptor
–Multi-query retrieval:
•multiSearchByFaceId, multiSearchByFaceDescriptor
–
10/11

Jan Sedmidubsky
April 7, 2014
Face Recognition Technology
GUI Design
•
•
•
•Thank Petra and her husband very much:
•http://www.fi.muni.cz/~xkohout7/facematch/index.html
•
11/11