Handwriting Copybook Style Analysis Of Pseudo-Online Data

1
Handwriting Copybook Style Analysis Of
Pseudo-Online Data

• Student and Faculty Research Day
• Mary L. Manfredi, Dr. Sung-Hyuk Cha, Dr. Charles
  Tappert, Dr. Sungsoo Yoon
• May 6, 2005

2
Sample of Copybook Style
3
Common Forensic Problem Determine Writer of
Questioned Document e.g., Ransom Note
4
Feasibility Study that might help in the writer
identification process

• Reduce the suspect population of the writer of a
  questioned document
• Identify the Copybook Style of a questioned
  document
• Use cluster analysis to reveal similarities among
  copybook styles

5
I/O Sample

• Input
• Output
• Copybook style usa2

6
Many Techniques Used for Writer Identification
(Online and Offline)

• Fragmented Connected-Component Contours
• Hidden Markov Methods
• Neural Networks
• Gray Level Distribution
• Fractal Analysis
• Directional Element Features and Linear Transform
• All techniques have pros and limitations
• More accurate ways always being sought

7
Offline versus Online Data

• Offline data are usually scanned images and only
  have static information
• Online uses special equipment (e.g., tablet
  digitizers, pen computers) to also capture the
  dynamic information (stroke number, order,
  direction, pressure, velocity etc..) as a person
  writes

8
Pseudo-Online Data

• Taking offline data and tracing it using a mouse
  or pen-enabled tablet to give it dynamic
  characteristics.

9
Offline Feature extraction
Distances between characters using Template
Matching
10
Pseudo-Online Feature extraction
Distances between characters using SDSS (stroke
direction sequence strings)
11
Stroke Directions
12
DB Creation

• Obtain various Roman copybook styles both cursive
  and manuscript
• Scan in each letter
• Apply normalization procedures to the letters
• Trace each letter with either a mouse or
  pen-enabled tablet and capture its dynamic
  characteristics

13
Methodology for processing a questioned document

• Determine Copybook Style
• Trace each letter of interest from a questioned
  document
• Match each letter against each of the
  corresponding letters in the DB. The letter with
  the smallest distance is the matched copybook
  style

14
Questioned Document
15
Distance Matrix (try to determine the questioned
document style)
16
Further analysis of Copybook Styles

• Clustering
• It is the unsupervised organization of patterns
  into groups so that the patterns in the same
  group are more similar to each other than to
  patterns in other groups
• Bring most similar letters together based on
  certain characteristics (here it is distance)
• We take each letter in the copybook style DB and
  calculate the distance between all others of the
  same letter (compare all uppercase As against all
  other uppercase As )

17
Partial Distance Matrix for Cluster Analysis
18
Dendrogram of Uppercase As
19
Clusters Resulting from the Uppercase A copybook
style
20
Summary

• Feasibility study for a new approach to help
  writer identification by narrowing the suspect
  population by
• determining the copybook style of a questioned
  document
• Cluster analysis of the copybook styles
• Will it help ? More work necessary to determine
  this

21
Future Work

• Do a more extensive analysis of the copybook
  styles. This would include adding more copybook
  styles to the DB, producing and naming the
  resulting clusters and producing various copybook
  style classifications.
• Get input from document examiners as to the
  possible value of this work