a computational tool for questioned handwriting … · a computational tool for questioned ... •...
TRANSCRIPT
CEDAR-FOXA Computational Tool for Questioned
Handwriting Examination
Computational Forensics
• Forensic domains involving pattern matching• Motivated by Importance of Quantitative methods in the
Forensic Sciences
1. Daubert Ruling2. High Standards established by DNA3. Computers
1. Low Cost2. Advances in Artificial Intelligence/Pattern Recognition
4. Improved Statistical Methods for EvidenceE.g., Aitken and Taroni, Statistics and the Evaluation of Evidence for Forensic Scientists, Wiley, 2004
• Bureau of Justice Statistics (2002)- Among 50 largest publicly funded crime labs
* 57% perform QD function* 5,231 cases requested* 1,079 backlogged at year end
• Significantly larger case load internationally
• Handwriting is common in QD case work
QDE
CEDAR Research on Handwriting QDE• Quantifying discriminatory power of handwriting
- Testing on national database, twins data
• Feedback from QDE’s in developing computational tools- Workshops at ASQDE- JtMtg of MAFS, CAFS- SWAFDE
• Developing Statistical Evidence Theory
CEDAR-FOX Software SystemPrincipal Functions
• Writer Verification/Identification• Document Properties• Signature Verification• Document Search
Computer System Requirements• Processor
– Pentium class processor– P4 or higher recommended
• Operating Systems– Windows NT, 2000, XP, Vista
• Random Access Memory– 256MB on XP and earlier– 512MB on Vista
• Secondary Storage– 30MB available disk space
Writer Verification
Known
Questioned
Sample Preparation: Rule Line RemovalOriginal Ruled Text
User Control
Removed Lines
Associating Truth with Word Images
Image
Truth/Transcript
TranscriptMap
Transcript Mapping
Extracted Characters (Letters)
Features Extracted
Distance and LLR Value
Distance = 0.35LLR = -0.26
Distance= 0.16LLR = 1.49
Distance = 0.43LLR = -0.97
Histograms and PDFs of Distances
0 5 10 15 20 25 30 35 400
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Distance
Pro
babi
lity
dens
ity
Same writerDifferent writer
0 5 10 15 20 25 30 35 400
20
40
60
80
100
120
Cou
nt
Same writer
0 5 10 15 20 25 30 35 400
50
100
150
Distance
Cou
nt
Different writer
0 0.1 0.2 0.3 0.4 0.5 0.6 0.70
0.5
1
1.5
2x 10
4
Cou
nt
Distance
Same writer
0 0.1 0.2 0.3 0.4 0.5 0.6 0.70
0.5
1
1.5
2x 10
4
Distance
Cou
nt
Different writer
0 0.1 0.2 0.3 0.4 0.5 0.6 0.70
1
2
3
4
5
6
Distance
Pro
babi
lity
dens
ity
Same writerDifferent writer
Macrofeature: Slant Microfeature: Letter “e”
SameWriter
DifferentWriter
Comparison of Words
Distance = 0.3702LLR = -0.35
Distance = 0.2022LLR = 4.44
Word Shape Comparison
Bigram Shapes
Distance = 0.1996LLR = 4.35
Distance = 0.3735LLR = -0.47
“th” combination and similarity score
Comparing Letter Pairs
Macro Features
Macro and Micro Feature Scores
Pictorial Attribute Scores (Macro)
LetterFormationScores (Micro)
Results of VerificationFeature Comparison Table
Strength of Evidence
Strength of Evidence Computation
• Based on similarities in a representative database of 1,500 writers providing 3 pages of writing each
• Probability distributions of similarities modeled by Gamma and Gaussian distributions
Similar Writing of Twins
LLR = 7.15
Dissimilar Writing of Twins
LLR = -98.36
Ranked Document List
Writer Identification
Word Recognition
Lexicon Selection
Word ComparisonAnd Similarity Score
Word Similarities
Document Properties
Document Line Structure
Document Properties
User selectsCharacter to be displayed
Comparing Letter Formations
Contour Display
Query Image
Searching Documents by Word Image
Searching Documents by Text Query
Retrieval: Word Images Retrieval: Words (Text)Retrieval: Word Images
Query: Text Word Query: Word Image Query: Word Image
Search Modalities
Genuine Set Scores forQuestionedSignatures
Signature Matching
AvailableIn HelpMenu
Organized by TopicsHierarchically
User Manual
Tool Bar Icons
• CEDAR-FOX is a system for assisting the QDE in dealing with handwriting
• Has automated tools for writer/signature verification/identification
• Has tools for case-work display• Computes strength of evidence
Summary
• Current system does not replace human examiner– It is a tool that can assist the examiner by
providing quantitative measures• Relationship to Frye, Daubert, Kumho
– Method produces reproducible results• Under same inputs
– Error rates can be established using system on large number of cases
Case Work Application
• Due to many functions in CEDAR-FOX it is necessary to gain familiarity with its use
• No formal training program set up yet
Competency Training
• Has been tested by several agencies:– Canada Border Agency– FBI with results presented at ASQDE-Montreal– USSS internal testing– Trial versions with several QDEs
• Further feedback solicited
QD Community Acceptance
• To be included in CEDAR-FOX version 1.2– Additional Tools for Image Manipulation
• Eraser Tool– Database Interfaces
• MySQL
Upgrades to CEDAR-FOX
• Improved Statistical Model– Current statistical model in system uses
independence assumption– Performance is not high as with better theoretical
models, e.g., neural networks– Plan to incorporate a compromise model e.g.,
pairwise independence
Upgrades to Software: Future Releases
Future Releases: Line Segmentation Improvements
Thank You