Terrorists

1
Terrorists

• Face recognition of suspicious and (in most
  cases) evil homo-sapiens

2
The problem

• Terrorists need to be identified when passing a
  security screen
• Aim is positive identification of a few faces
• Problem is that terrorists try to disguise
  themselves

3
About Us

• Team members
• Gülsah Tümüklü (manager)
• Réka Juhász
• Emil Szimjanovszki
• Gergely Windisch

4
Goal

• Finding the terrorists
• Identifying faces even if they are disguised

5
System
6
Our Implementation

• Programmed in Matlab
• Input RGB image
• Pre-processing output Greyscale image
• Output Yes/No (Terrorist-wise)

7
Things We Do

• Acquire an Image

8
Things We Do (2)

• Locate eyes

9
Things We Do (3)

• Normalise (rotate, scale, clip, put eyes to their
  place)
• 128128

10
Things We Do (4)

• Face Recognition (details later)

11
Things We Do (n)

• Decide, then Call the police

12
101 Useful Tips for Terrorists
13
101 Useful Tips for Terrorists
14
101 Useful Tips for Terrorists
15
101 Useful Tips for Terrorists
16
101 Useful Tips for Terrorists
17
Recognition Part

• Problem Face Recognition
• Literature about Face Recognition
• Problems in Face Recognition
• Eigenfaces

18
Problem Face Recognition

• Identifying persons using some priori information
• Many potential applications, such as person
  identification, human-computer interaction,
  security systems, image retriveal systems, and
  finding terrorists ?
• Stages of Face Recognition
• face detection
• feature extraction
• facial image classification

19
Literature about Face Recognition

• Classification of Face Recognition Methods
• Hollistic Methods
• Feature-Based Methods
• Hybrid Methods

20
Face Recognition Methods

• Hollistic Methods
• PCA
• Eigenfaces, Probabilistic eigenfaces,
  Fisherfaces/subspace LDA , SVM, Evolution
  pursuit, Feature lines, ICA
• Other Representations
• LDA/FLD, PDBNN

21
Face Recognition Methods(cont)

• Feature-Based Methods
• Pure geometry methods
• Dynamic link architecture
• Hidden Markov model
• Convolution Neural Network

22
Face Recognition Methods (cont.)

• Hybrid Methods
• Modular eigenfaces
• Hybrid LFA
• Shape-normalized
• Component-based

23
Problems in Face Recognition

• Feature Extraction
• Global Features
• Local Features
• Handling some problems
• Illumination differences
• Facial expressions
• Occlusions
• pose

24
Eigenfaces

• Firstly introduced by Pentland, and Turk in 1991
• It is considered the first working facial
  recognition technology
• Based on PCA
• Decompose face images into a small set of
  characteristic feature images called eigenfaces
• Eigenfaces may be thought of as the principal
  components of the original images

25
Eigenfaces (cont.)

• Trainning Part calculate the Eigenfaces of
  datases
• Classification part Reconstruct the test image
  and classify it

26
Calculation of Eigenfaces

• Calculate average face v.
• Collect difference between training images and
  average face in matrix A (M by N), where M is the
  number of pixels and N is the number of images.
• The eigenvectors of covariance matrix C (M by M)
  give the eigenfaces. M is usually big, so this
  process would be time consuming.

27
Calculation of Eigenfaces(cont.)

• Use SVD
• Substract mean image from training images
• diff.imagestrainingimages-mean image
• Find the svd of diff.images
• U S V svd(diff.images)
• The columns of U are automatically the e-vectors
  of diff.images diff.images
• Square of S gives eigenvalues

28
(No Transcript)
29
Classifying a test Image

• Find the reconstructed image
• Calculate weights
• First find difference test image
• Diff.testtest Image-mean Image
• Do inner product of each eigenimage with the
  difference image to get a weight vector
• Find the reconstructed image
• for m1numTrainingImages
• reconstructionImage reconstructionImage(
  weight(m)Eimage(,,m)) end

30
Classifying a test Image (cont.)

• If one of weighs is above a threshold, take the
  largest one and return that its owner also owns
  the new face.
• Use nearest neighbor method
• Find minumum distance between reconstructed image
  and eigenfaces and assign test image to class
  which has min distance

31
Pros and Cons

• Pros
• It is fast
• Efficiency
• Provides accurate recognition rates
• Cons
• Very sensitive to occlusions, illuminations,
  facial expression, pose
• Only works good with frontal faces

32
Results (1) Training set
Class 1 (Terrorists)
Class 2
Class 3
33
Results (2)
34
Conclusion

• Face recognition is a difficult problem
• Pre-processing is very important
• It is not enough to use only global features
• Better results can be obtained with different
  classifications (eigenfeatures)

35
References

• M. Turk and A. Pentland. Eigenfaces for
  recognition. Journal of Cognitive Neuroscience, 3
  (1), 1991a.
• M. A. Turk and A. P. Pentland. Face recognition
  using eigenfaces. In Proc. of Computer Vision and
  Pattern Recognition, pages 586-591. IEEE, June
  1991b.
• W. Zhao, R. Chellappa, P. J. Phillips and A.
  Rosenfeld, "Face Recognition   A Literature
  Survey", ACM Computing Surveys(CSUR), vol. 35,
  issue 4, pp. 399-458, December 2003.
• http//cilek.ceng.metu.edu.tr/facedetect
• B. Galamb Color Based Eye Location