Face Detection - PowerPoint

							Face Detection

EE368 Final Project

       Group 14
     Ping Hsin Lee
   Vivek Srinivasan
  Arvind Sundararajan
                    Overview

 Introduction
 Methods used to detect faces
     Color segmentation
     Morphological Processing
     Template Matching And Clustering
     Results
 Techniques considered but not used
              Color Segmentation

 Use color information in the YCbCr domain
  •YCbCr Color space
  effectively decorrelates
  the intensity and color
  information

  •Each channel
  information is represented
  in discrete levels.
                            MAP Rule
 Implement MAP decoder to determine skin from non-
  skin pixels
 D(I(x,y)) = 1 if P(I(x,y) | S)P(S) > T* P(I(x,y) | NS)P(NS)
            = 0 other wise

  •Minimizes
  misclassification error
Result of Color Segmentation
      Morphological Processing

Reject blobs of small sizes, perform closing, remove holes
         Non-face Object Removal

 Use information about
  shape and location of
  objects in conjunction to
  reject non-face objects
  while minimizing rejection
  of faces
 Objects characterized by
  max/min as a measure of
  length. (Independent of
  size, translation, and
  rotation of objects)         Example of non-face
                               object removed by CCA
Non-face Object Removal

   Before and after rejection
                Template Matching

 Performed in the              Rejected Template
  luminance domain using
  the FFT
 First attempt: use the
  average of all face regions
 Features did not seem to
  align properly, hence this
  template was rejected
          Final Templates Used
a)                 Resample each face region to the
                    same size before averaging.
                   Include mirror images of each
                    face region to produce a
                    symmetric template (a).
                   In addition, a non-symmetric
                    partial template (b) is used to
                    capture information about
     b)             smaller and partially obscured
                    faces in the image
                   One template tests for symmetry,
                    while the other tests for non-
                    uniform illumination, and
                    captures smaller faces as well.
    Clustering of Correlation Peaks

 The autocorrelation results for each template were
  first thresholded and then combined.
 Used heuristic techniques based on shape of the
  skin regions to group peaks.
 Any 2 peaks meeting a maximum distance criterion
  and connected by a line passing through only skin
  regions were grouped together as a face.
      Results of Grouping
       Correlation Peaks
     Before and after peak grouping

a)                   b)
Results Applied to the Original
           Image
Image corresponding to the grouped peaks
     c)
                  Final Results


Image   1    2       3   4    5    6    7

Score   20   22     24   22   24   24   21

Total   21   24     25   24   24   24   22
      Techniques Considered but not
                  Used
 Fisher’s linear discriminant (FLD)
     Poor performance in rejection of false positives
      because detected non-face and face regions are
      not linearly separable
 Eigenfaces
     Produced results similar to template matching
      but at an increased computational cost
      Techniques Considered but not
                  Used
 Support Vector Machines (SVM)
     Generated 470 face regions and 500 non-face
      regions each of size 49x55 pixels as training
      database
     Employed a Gaussian radial basis function (RBF)
      as kernel
     Samples of database images

 Faces



 Non-faces
              Results of SVM

 Produced decision regions that are too tightly
  bound to the training face samples and were
  not able to classify the faces in the other
  training pictures
 Including the SVM in the program would
  only slow down our runtime and would not
  produce noticeable improvements
                    Conclusion

 Color segmentation in the YCbCr domain and
  morphological processing produced good estimates
  of face regions
 Implemented multi-resolution template matching
  and peak clustering to further distinguish different
  face regions from each other and from non-face
  regions
 Could have done more to reject false positives
  (MRC/neural networks to reject hand regions)
     Face Detection   Gender Recognition
1        4 (19)             2 (1)
2        2 (20)             5 (0)
3        8 (18)             5 (0)
4        8 (18)             5 (0)
5        13 (13)            5 (0)
6        10 (17)            5 (0)
7        11 (16)            5 (0)
8        12 (14)            1 (2)
9        1 (22)             2 (1)
10       2 (20)             2 (1)
11       14 (08)            5 (0)
12       4 (19)             5 (0)
13       4 (19)             5 (0)
14       4 (19)             5 (0)
15
16
17