幻灯片 1 CPSC 601 Lecture Week 5 Hand

Document Sample
幻灯片 1 CPSC 601 Lecture Week 5 Hand Powered By Docstoc
					CPSC 601 Lecture Week 5

    Hand Geometry
Outline:
1.   Hand Geometry as Biometrics
2.   Methods Used for Recognition
3.   Illustrations and Examples
4.   Some Useful Links
5.   References
Hand Geometry

Hand geometry is a biometric technique,
 which identifies person through the hand
 geometry measurements.
 Some geometric structures related to a
 human hand (e.g., length and width of
 hand) are relatively invariant to an
 individual.
Hand Geometry
Characterized by its lengths, widths,
  shapes etc.
Advantages:
  (1) Acquisition convenience and good
  verification performance
  (2) Suitable for medium and low security
  applications
  (3) Ease of Integration
Disadvantages:
  (1)Large size of hand geometry devices
  (2)Only used for verification
  (3)Single hand use only
                                        Picture taken from:
                                        http://www.handreader.com/
Hand Geometry

Properties:
  Medium cost as it need a platform and a medium
   resolution CCD camera.
  It use low-computational cost algorithms, which leads to
   fast results.
  Low template size: from 9-25 bytes, which reduces the
   storage needs.
  Very easy and attractive to users: leading to nearly null
   user rejection.
Application

 Hand geometry information is not very distinctive
 The hand based biometric systems can be
  employed in those applications which don't
  require extreme security but where robustness
  and low-cost are primary issues.
Comparison between biometrics
Hand Geometry Verification System
Hand Geometry Biometric System

Biometrics System
  Image Processing
  Feature Extraction
  Feature Matching
System demonstration




         Hand Geometry Verification System
System demonstration
 Hand Subsystem




      A flowchart for hand feature extraction and matching
Binarization
 (1) Change input RGB image into gray-level image
 (2) Change the gray-level image into white-black image.
 (3) Due to illumination problems, Median filtering to
remove noise is used. G(i,j) represents the gray value of pixel (i,j) after
    binarization, I(i,j) represent the original gray value.




                  1, if I (i, j )  threshold
                  
      G (i, j )  
                  0, otherwise
                  
Binarization Results


(a) Input Image (b)Gray-Scale   (c) Before filtering   (d)After filtering
Border Tracing
(1) Searching for the starting point
(2) Use the following algorithm




(3) All the coordinates of the border are recorded
Border Tracing
 (a) Binary Hand   (b) Hand Contour
Point Extraction
Purpose: To pinpoint the five finger tips and four finger roots.
Method: Depict the vertical coordinates of all contour pixels
Points Extraction
By computing the first-order differential of vertical coordinates of f(i),
mark where differential sign changing from -1 to 1 as finger tips,
where differential sign changing from 1 to -1 as finger roots.
Measurement
Generate a feature vector Vh, including 5 lengths of fingers, 10 widths
of fingers, and the width between v1 to v2.
Feature Vector Matching
 Let F = (f1; f2; :::; fd) represent the d-dimensional feature
  vector in the database associated with the claimed
  identity and Y = (y1; y2; :::; yd) be the feature vector of
  the hand whose identity has to be verified.
 The verification is positive if the distance between F and
  Y is less than a threshold value. Distance metrics,
  absolute, weighted absolute, Euclidean are used to
  compute distance.
Distance Matrices Matching
                                    N

                                  | y  f
Absolute distance metric
                                               i      i   |
                                   i 1
Weighted absolute metric          N
                               i 1
                                         (( yi  f i ) /  i ) 2

Euclidean distance metric
                                          N
                                       i 1
                                                     ( yi  f i ) 2
Other Feature Matching Algorithms

 Hamming Distance
   This distance doesn’t measure the difference
    between components of the feature vectors, but
    the number of components that differ in value.




   # {i {1,..., N} / | y i  f i | threshold i }
Other Feature Matching Algorithms
 Gaussian Mixture Models
   This is a pattern recognition technique that uses an
     approach between the statistical methods and the
     neural networks. It is based on modeling the patterns
     with a determined number of Gaussian distributions,
     giving the probability of the sample belonging to that
     class or not. The probability density of a sample
     belonging to a class u is:


               N

                (2 )                                             
                              ci                   1   T            1      
p( x / u )                                    exp{ ( x  u i )            ( x  u i )}
               i 1
                         L/2
                               |   i
                                        1/ 2
                                        |           2              i
Other Feature Matching Algorithms

Radial Basis Function Neural Networks
  A neural networks method. First train the net
   using a set of feature vectors from all the users
   enrolled in the system, and each output will
   correspond to each class. Then, the new
   feature vector is inputted into the net, and
   classified as one of the class in the database.
Performance Evaluation
 FAR and FRR stands for false acceptance rate
  and false rejection rate, respectively. The FAR
  and FRR are defined as below:




 Equal error rate (EER) where FAR = FRR.
Image Acquisition




(a) Hand geometry sensing device   (b) Incorrect placement of hand
A Typical System

Hand Shape Identification System
 (Biometric Systems Lab, University of
 Bologna, Italy.) extracts 17 geometric
 features from the hand ( finger length and
 widths, hand width and height, ...).
A Typical System
A Typical System
 The experimental studies on a sample of 800
  images (100 people, 8 images for each one)
 The main characteristics of HaSIS are as follows:
  FAR = 0.57 %
  FRR = 0.68 %
  verification time = 0.5 sec.
  enrollment time = 1.5 sec.
Access Control through Hand Geometry
(Purdue Univ.)
Useful Links
 Biometric Systems Lab, Univ. of Bologna, Italy.
  http://bias.csr.unibo.it/
 Biometric Research Center, Hong Kong
  http://www4.comp.polyu.edu.hk/~biometrics/
 Biometric Lab, Purdue Univ.
  http://www.tech.purdue.edu/it/resources/biometrics/
 Biometric Research, MSU
  http://biometrics.cse.msu.edu
References

 Goh Kah Ong Michael, AUTOMATED HAND GEOMETRY VERIFICATION
  SYSTEM BASE ON SALIENT POINTS. The 3rd ISCIT.
 Arun Ross, A Prototype Hand Geometry-based Verification System, IEEE Trans.
  PAMI, vol. 19, no7.
 Paul,S etc, Biometric Identification through Hand Geometry Measurements.
  IEEE Trans, PAMI Vol 22, No. 10,2002


				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:3
posted:12/8/2011
language:
pages:31