Docstoc

Validation of Individuality of Handwriting _Dichotomy Model_

Document Sample
Validation of Individuality of Handwriting _Dichotomy Model_ Powered By Docstoc
					                                                          CSIS


 The Interplay of Student Projects
  and Student-Faculty Research

     Charles C. Tappert, Allen Stix, and Sung-Hyuk Cha
Seidenberg School of Computer Science and Information Systems




                         E-Learn 2007
                                                CSIS

               Third Paper on
         Real-World Student Projects

1. Integrating Real-World Projects for Actual
   Customers into Capstone Courses (E-Learn 2002)
2. Security-Related Real-World Projects (ISECON
   2004)
3. Interplay of Student Projects and Student-Faculty
   Research (E-Learn 2007)


                       E-Learn 2007
                                               CSIS

      Research and Project Interplay
• Research can require supporting infrastructure
• We create projects to develop that
  infrastructure for student/faculty research
• Typical infrastructure systems
  –   Client/server systems
  –   PC based systems
  –   Databases
  –   Web interfaces
  –   Networking systems
                         E-Learn 2007
                                                               CSIS

        Real-World Projects 2001-2007
• Total of 63 Projects with 183 Students
   – Average of about 3 students per project
• Involved Courses – Mostly Masters Level
   – M.S. Capstone Courses – 52 Projects
      • CS615-616 Software Engineering
      • IT691-Capstone Project and CS691-CS Projects
   – M.S. Elective Courses – 7 Projects
      • Pervasive Computing
   – Undergraduate Courses – 4 Projects
• Students Build Systems for Real Customers
   – Internal University Community – various schools/departments
   – External Community – local organizations and other universities
                               E-Learn 2007
                                                         CSIS

Summary of Projects and Publications
                                      Project
                   Number Project             Offshoot
  Project Category                    Related
                   Projects Semesters          Pubs
                                       Pubs
 Web Applications          8              12   8
 Pervasive Systems         14             24   18
 PC Applications           10             17   11
 Artificial Intelligence   6              8    8
 Pattern Recognition       8              11   27   19
 Biometric Systems         12             15   17   19
 Quality Assurance         5               9   5
 Totals                    63             96   94   38
                           E-Learn 2007
                                        CSIS

          Project Sources
      Project Source           Number
Faculty Ideas or Research        32
Student Ideas or Research        13
External Community               10
Internal University Needs        8
Totals                           63

                E-Learn 2007
                                           CSIS

         Publication Types
     Publication Type             Number
External Conference Papers          48
Journal Articles                    4
Book Chapters                       1
Doctoral Dissertations              15
Masters Theses                      3
Internal Conference Papers          57
Internal Technical Reports          4
Totals                             132
                   E-Learn 2007
                                                                           CSIS

      Project Categories and Examples
• Web Applications
   – Online Course Survey, Genealogy Web App, Weather Info System
• Pervasive Systems
   – VoiceXML Apps, Hospice Nurse Handheld, Medical Vital Sign Wearable
• PC Applications
   – Cluster/Grid Systems, Music Pitch Training Sys, Astronomy Image Database
• Artificial Intelligence
   – Bridge Bidding System, Mixed-Reality Pocket Billiards System
• Pattern Recognition
   – Interactive Visual Systems: Flowers, Rare Coins, Flags, Pottery, Paintings
• Biometric Systems
   – Iris, Face, Handwriting, Voice, Keystroke, Mouse Movement, Stylometry
• Quality Assurance

                                 E-Learn 2007
                                                       CSIS

                 Student Research
• Original, rigorous work that
   – advances knowledge
   – improves professional practice, and/or
   – contributes to the understanding of a subject
• Methods depend upon the nature of the research:
   –   controlled experiment
   –   empirical studies
   –   theoretical analyses
   –   other methods
• We require research work to be of sufficient strength
  so that students can distill from it a paper worthy of
  publication
                               E-Learn 2007
                                                        CSIS

    Doctor of Professional Studies in
     Computing (D.P.S.) Program
• Initiated in 1999
• First graduates in 2002
• Enables computing and IT professionals
  – To earn a doctorate in three years
  – Through part-time study while continuing in their
    professional careers



                       E-Learn 2007
                                                CSIS

              Student Projects
• Students develop computer information
  systems for actual customers
  – Many projects support ongoing research
• Students use known technology to develop
  systems according to specified requirements
• Project systems serve the community
  – internal university community at Pace
  – greater university community
  – external non-profit local community

                       E-Learn 2007
                                                           CSIS

     Project Customer Affiliations
• Pace University
  – Various schools/departments: School of Computer Science
    and Info Systems, Business School, Nursing School, Arts
    and Sciences, Dept of IT, Assessment Office
  – Researchers: faculty and dissertation students
• Outside Organizations
  – Universities: Rensselaer Polytechnic Institute, SUNY New
    Paltz, Yonsei University in Korea, SUNY Buffalo
  – Non-profit: Northern Westchester Hospital, Columbia
    Presbyterian Medical Center, Phelps Memorial Hospital
  – Research Institutions: IBM T.J. Watson Research Center

                         E-Learn 2007
                                                CSIS

       Student Project Team Make-up

•   Architect-Designer
•   One or two Implementers
•   Quality Officer
•   Coordinator-Liaison (usually team leader)

    (Note: several team member functions
     can be combined for small teams )

                      E-Learn 2007
                                               CSIS

 Project Development Environment
• Development Servers
  – Pentium NT servers
  – Solaris Unix servers
• Software
  – Database software
     • MySQL, Microsoft Access, Oracle
  – Scripting software
     • Active Server Pages, Cold Fusion, PHP
  – Tomcat for Java servlets, etc.
                         E-Learn 2007
                                                                            CSIS

Project-Research Interplay Examples
• Handwriting Forgery Database/Quiz System
   – Collaborative project work with IBM Research Center
   – Work extended into a completed M.S. dissertation
   – Resulted in student/faculty conference and journal publications
• Interactive Visual Pattern Recognition System
   – Initial project was collaborative work with RPI on flower identification
   – Extension to flag identification resulted in a completed M.S. dissertation and
     student/faculty conference publications
   – Rare Coin Grading System resulted in completed doctoral dissertation
   – Extensions to Pottery and Painting led to two additional doctoral dissertations
• Keystroke Biometric Database and Testing System
   – Sequence of several projects led to two completed doctoral dissertations
   – A third doctoral dissertation is in progress on a related problem

                                 E-Learn 2007
                                           CSIS

Links to Example Project Systems

• Handwriting Forgery Quiz System
• Rare Coin Grading System
• Keystroke Biometric Information System




                   E-Learn 2007
        Student and Faculty Research/Projects                 CSIS


• Studies in Biometric Identification/Authentication
   – Iris
   – Face
   – Handwriting style and forgery detection
   – Voice
   – Keystroke, mouse movement, and stylometry
• Interactive Visual Systems (initially in collaboration with RPI)
• Handwriting and Forensic Document Analysis (with IBM)
• Speech/Voice Related Studies (some with IBM)
• Object Tracking (Surveillance) System
• Wearable/Mobile/Pervasive Computing Research


                           E-Learn 2007
               Biometric Authentication                      CSIS




A robot identifies a suspect, from the movie “Minority Report.”
                          E-Learn 2007
                     Iris Authentication: Data   CSIS

              Left   Right




      Train


Man


      Test




      Train



Wo
man
      Test




                             E-Learn 2007
Iris Authentication: Image Processing   CSIS




              E-Learn 2007
                Iris Authentication: System                                              CSIS




                                 Feature extraction

    ( f1x , f 2x ,..., f dx )                           ( f1 y , f 2y ,..., f dy )

                                Distance computations



 ( f1 x , f 1 y )    ( f 2x , f 2y )                                ( f dx , f dy )




                                     Dichotomizer


                                Same/different people
                                         E-Learn 2007
     Face Recognition            CSIS




Each person has a unique face?
          E-Learn 2007
        Face Recognition: System             CSIS




              ?
Query




                                   Face DB
               E-Learn 2007
Inspirational Portrait of Individuality   CSIS




               E-Learn 2007
Face Recognition: National Security   CSIS




             E-Learn 2007
        Individuality of Handwriting: Legal Motivation                               CSIS



                              To determine the Validity of
                              Individuality in Handwriting



                                                    Daubert vs. Merrell Dow
               Frye vs. US (1923)                        (1993) testing,
              scientific community                   peer review, error rates



U.S. vs. Starzecpyzel            GE vs. Joiner                      Kumho vs.Carmichael
        (1995)                       (1997)                                 (1999)
 “skilled” testimony            weight of evidence                   reliability standard




                                     E-Learn 2007
Ted Bundy Document   CSIS




      E-Learn 2007
Individuality of Handwriting: Data     CSIS




     Each person writes differently.
              E-Learn 2007
              Handwriting Analysis Taxonomy                               CSIS


                       Analysis of Handwriting




    Recognition             Examination              Personality identification
                                                          (Graphology)


On-line     Off-line   Writer Identification     Writer Verification



                          Natural Writing      Forgery      Disguised Writing




                              E-Learn 2007
    Handwriting: Genuine vs. Forgery              CSIS




(a) Genuine handwriting samples from one writer




              (b) Forgeries of (a)
                  E-Learn 2007
     Three Differences between Genuine & Forgery   CSIS


1. Shape




2. Pressure




3. Speed




                      E-Learn 2007
                               Automatic Forgery Detection Model                                                                              CSIS
sample1 by x                           sample2 by x                            sample1 by x                           Forgery of x by y




                                                       Feature Extractor
                                                                                                                             x       x             x
( f1x1 , f 2x1 ,..., f dx1 )         ( f1x2 , f 2x2 ,..., f dx2 )              ( f1x1 , f 2x1 ,..., f dx1 )              ( f1 y , f 2 y ,..., f d y )

                                                    Distance computing

 (d ( f1x1 , f1x2 ), d ( f 2x1 , f 2x2 ),..., d ( f dx1 , f dx2 ))                                x                  x                         x
                                                                                   (d ( f1x1 , f1 y ), d ( f 2x1 , f 2 y ),...,d ( f dx1 , f d y ))

                d-dimensional                                                                d-dimensional
               within-authentic-                                                           between-authentic-
                 handwriting                                                             handwriting & forgery
                 distance set                                                                 distance set

                                                                    E-Learn 2007
                                Artificial Neural Network                    CSIS

Authentic
sample from a                                          ( f1 x , f 1 y )
known source
                       x    x
                 ( f1 , f ,..., f )    x               ( f 2x , f 2y )
                           2          d



                                      Distance
           Feature                                                         Original/
                                      compu-
          extraction                                                       Forgery?
                                       tation

                 ( f1 y , f 2y ,..., f dy )



Handwriting                                            ( f dx , f dy )
sample in
question

                                              E-Learn 2007
     Detection of Forgery by Novices: Hypotheses CSIS


• Good forgeries – those that retain the shape and
  size of authentic writing – tend to be written
  more slowly (carefully) than authentic writing

• Good forgeries are likely to be wrinklier (less
  smooth) than authentic handwriting




                      E-Learn 2007
     Forgery Detection: Methodology                  CSIS

•   Sample collection: online, scan to get offline
•   Feature extraction: Speed, Wrinkliness
•   Statistical analysis




                    E-Learn 2007
Fractal Measure: How Long is a Coastline?   CSIS




               E-Learn 2007
Fractal: How wrinkly is the Coastline of Britain?   CSIS




                   E-Learn 2007
       Fractal: How wrinkly is Handwriting?          CSIS




           (a)                           (b)
        (a) Number of       in the boundary = 69
        (b) Number of        in the boundary = 32
                  boundary _ in _ high _ res. 
                        69
Wrinkliness  boundary _ inlog(2) . 1.1085
Wrinkliness  log log(
                           ) / _ low _ res  / log(2)
                                               
                       32                     

                        E-Learn 2007
      Detection of Forgery by Novices: Results   CSIS



• Significance results (T-test)
   – Forgeries are written slower: p = 5.90E-09
   – Forgeries are wrinklier: p = 0.0205
• Importance
   – We can detect the wrinkliness, and
     therefore detect forgeries, from offline
     data, i.e. from scanned images

                     E-Learn 2007
Speaker Individuality: “My name is …”   CSIS




              E-Learn 2007
            Speaker Individuality      CSIS

“My name is” from Two Different Speakers




                  E-Learn 2007
             Speaker Individuality         CSIS




“My name is” divided into seven sound units.




                   E-Learn 2007
   Speaker Individuality: System                                 CSIS

                                           ( f1 x , f 1 y )


   ( f1x , f 2x ,..., f dx )
                                           ( f 2x , f 2y )

Feature                   Distance
Extrac-                   compu-                                Same/
  tion                     tation                              Different


   ( f1 y , f 2y ,..., f dy )



                                           ( f dx , f dy )



             98 percent accuracy
                                E-Learn 2007
          Multi-modality Biometric Authentication                       CSIS
                                          System that requires
                                            user verification

                                                   Embeded & Hybrid
                                                    User Verification
                                                         system




                                                                  biomouse
LCD Pen                                      Digital
                   Microphone                                    Fingerprint
 tablet                                      Camera
                                                                   scanner
                           E-Learn 2007
Interactive Visual System (IVS)          CSIS


           We began this work in
           collaboration with RPI on a
           flower identification application

           However, IVS is a technology,
           not just a flower identification
           application

           We also have results on flag
           recognition, and we plan to
           explore the applications of sign,
           face, and skin-lesion recognition


          E-Learn 2007
         Interactive Visual System: Motivation          CSIS

• Image recognition can be a difficult problem

• Modern AI and pattern recognition techniques try to
  automate the process – that is, they do not include the
  human in the equation

• Humans and computers have different strengths
   – Computers excel at large memory and computation
   – Humans excel at segmentation

• We propose combining human and computer to
  increase the speed and accuracy of recognition

                         E-Learn 2007
       Interactive Visual System: Flower User Interface   CSIS



•   Load Flower Image
•   Select Features
•   Identify
•   Previous 3 Hits
•   Next 3 Hits
•   Store New Flower
•   Auto Feature Extract
•   List Extracted Features



                         E-Learn 2007
Interactive Visual System: Flower Shape Model   CSIS




                 E-Learn 2007
      Interactive Visual System: Flag Recognition CSIS

We have extended the Interactive Visual System work to other
applications – e.g., we have preliminary results on flag recognition.




                            E-Learn 2007
      Interactive Visual System: Planned Applications   CSIS


• Foreign Sign Recognition
  – Shape model: rectangle

• Face Recognition
  – Shape model: 3D face template



• Skin Lesion Recognition
  – Shape model?


                        E-Learn 2007
           Benefits of Research and Project Interplay        CSIS

• Project students get real-world learning experience
   – Some students even get involved in research aspects
   – Promote interdisciplinary collaboration and local community
     involvement
• Student and faculty researchers benefit from project-
  created supporting infrastructures
   – Further student and faculty research
   – Increase number and quality of publications
• Enhance relationships between Pace University and other
  universities and technology companies
• Increase national recognition of our research centers and
  of the university
                           E-Learn 2007
                                     CSIS

         Questions/Comments?
• Dr. Chuck Tappert
  – Email: ctappert@pace.edu
• Dr. Allen Stix
  – Email: astix@pace.edu
• Dr. Sung-Hyuk Cha
  – Email: scha@pace.edu




                      E-Learn 2007

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:1
posted:4/30/2012
language:
pages:52