Face Recognition

1
Face Recognition

• using Direct Fractional Step Linear Discriminant
  Algorithm (DF-LDA)
• Mohammad Anwar
• Sumit Sampat
• Mohammad Murtuza

2
Progress

• Understanding of related material is completed.
• Beginning of implementation has begun.

3
DF-LDA Algorithm
(size of the image)
The output are the Optimal Discriminant Features
(ODF) which will be projected onto the DF-LDA
based subspace.
4
DF-LDA Algorithm contd.
between class scatter matrix and is where

• choice of weighting function - Euclidean

5
DF-LDA Algorithm contd.
gt 0
diagonal matrix.
6
DF-LDA Algorithm contd.
Since we need to maximize the ratio
7
DF-LDA Algorithm contd.
The basic idea of complex mathematical
computations in order To reduce the dimensions is
to obtain better classification. The F-LDA step
is incorporated in the case of closed classes to
obtain the required ODFs.
8
Feret Dataset

• Feret Dataset consists of facial images at
  different angles which are not normalized. Also
  includes data files which have feature data
  points (ground truths).
• Feature data points includes x, y-axis
  locations of facial features such as left and
  right eyes, nose, and center of mouth.

9
Feret Image Example
history FERET../00001fa010.930831 date_taken
31 Aug 1993 eye_glasses yes pose
frontal lighting inside_feret media film_still
ASA_200 expression FA id 00001 image_processing
histogram_adjusted image_resize
0 image_bright_reduction 0 feret_pose_flag
fa left_eye_coords 163 166 right_eye_coords 101
167 nose_tip_coords 134 202 mouth_center_coords
133 234
00001fa010_930831.tif
00001fa010_930831.gnd
10
Implementation preprocessing image files

• Each image in Feret dataset are bzipped and and
  in tiff image format.
• Using a simple batch script to unzip the files
  and convert tiff image files to PGM format using
  convert function from ImageMagick 5.5.7.
• URL for ImageMagick www.imagemagick.org

11
Implementation creating eye-coordinates file

• Each image in Feret has an associated .gnd file
  which lists feature data points.
• Perl script is used to extract left and right eye
  coordinates from each .gnd file and placed into
  an one eye coordinate file which lists image name
  and eye-coordinates.

00001fa010_930831.pgm 163 166 101
167 00002fa010_930831.pgm 135 136 100
189 00003fa010_930831.pgm 168 169 112 170
Eye coordinates file
12
Implementation normalization

• Normalization is done by reading each line in the
  eye coordinates file.
• These are the steps for normalization 4
• The image is scaled so as to make the distance
  between the eye's constant. In this step, the
  image is also cropped to a smaller size that will
  include essentially just the face. The standard
  FERET normalization crops the image to 150x130
  pixels with 70 pixels between the centers of the
  eyes.
• A mask is applied that zeroes out pixels not in
  an oval that contains the typical face. Thus,
  hair, shirt collars, etc. are typically removed.
  The mask is generated analytically by specifying
  the dimensions of the masking oval.
• Histogram equalization is used to smooth the
  distribution of grey values for the non-masked
  pixels.
• The image is normalized so the non-masked pixels
  have mean zero and standard deviation one.

00001fa010_930831.pgm
00001fa010_930831.nrm
13
Implementation Image Lists

• Image Lists is a file which lists on each line
  the same-class normalized images.
• Using Perl, we parse the directory list of
  normalized images and create this file.
• This image list is a reference for training to
  know which images belong to a single class.

SubjectA_Image1.nrm SubjectA_Image2.nrm
SubjectA_Image3.nrm SubjectB_Image1.nrm
SubjectB_Image2.nrm SubjectC_Image1.nrm
SubjectC_Image2.nrm SubjectC_Image3.nrm . . . .
image_list.srt
14
Implementation What's next?

• Need to implement the DF-LDA algorithm for
  training using normalized files.
• Experimentation and Analysis of results.

15
References

• Juwei Lu et. al. Face Recognition using LDA-Based
  Algorithms. IEEE Transactions of Neural Networks.
  Vol. 14, no. 1, January 2003.
• Feret Dataset. National Institute of Standards
  Technology.
• Peter Belhumeur et. Al. Eigenfaces vs
  Fisherfaces Recognition Using Class Specific
  Linear Projection. Vol 19. no 7. July 1997.
• Evaluation of Face Recognition Algorithms.
  Colorado State University. http//www.cs.colostate
  .edu/evalfacerec/data/normalization.html