Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Online Signature Feature Extraction
from Video
J´ozsef N´emeth
28 june 2012
J´ozsef N´emeth Online Signature Feature Extraction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents
Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
1 Introduction
2 The method Recording Tracking Comparison Classification
3 Results
Error measuring Results
4 Future work
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Introduction
Motivation
• Cheap, widely used device
• May contain more information
• Could be used with other devices
• Advantage: Real signature
Goals
• What kind of features can be extracted?
• Which feature (or feature set) gives the best results?
J´ozsef N´emeth Online Signature Feature Extraction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Video based verification
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Recording
J´ozsef N´emeth Online Signature Feature Extraction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Tracking
Pen tip tracking
• Intersection of lines (Hough transformation)
• Refinement using template matching
Pen angle
• Pen tracking using Hough transformation
Reconstruction
• Camera calibration using pattern
• Precalibrated camera could be used in a real application
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Reconstruction
J´ozsef N´emeth Online Signature Feature Extraction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Reconstruction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Pen angle
J´ozsef N´emeth Online Signature Feature Extraction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Features
Pen tip tracking
• Conventional features
• Coordinates→ velocity→ acceleration
• Normalization
Pen tracking
• Pen angular offset
• and its velocity
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Comparison of data sequences
Dynamic Time Warping (DTW)
• Measures the similarity between two sequences
• Data vary in time and/or speed
• Also provides a pairing
• Dynamic programming
J´ozsef N´emeth Online Signature Feature Extraction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
DTW
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Classification
The database
• 20 person
• 10-10 genuine signatures and 5-5 forgeries
Learning
• Learning set from 5 randomly chosen real signatures (T)
• The remaining signatures will be classified.
• Unknown signature (s)→ real or forgery?
J´ozsef N´emeth Online Signature Feature Extraction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Classification
Classification
• Average distance between the learning signatures (Davg(T, T))
• Average distance between the learning signatures and the unknown (Davg(T, s))
• The signature is accepted if
Davg(T, s)< t·Davg(T, T) wheret is a fixed constant.
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Error measuring
Error types
FAR - false acceptance rate FRR - false rejection rate
EER - equal error rate, where FAR and FRR are equal
J´ozsef N´emeth Online Signature Feature Extraction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Error measuring
Equal error rate
• Increasing the tthreshold: FRR decreasing, while FAR increasing
• Looking for t, where FRR=FAR⇒ EER
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Error measuring
Equal error rate
• How EER changes as we increase the elements of the database?
J´ozsef N´emeth Online Signature Feature Extraction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Results
Simple features
feature EER
point-coordinates 11.1%
velocity 14.4%
acceleration 13.0%
angle 6.0%
Pen angle 17.7%
Compound features
feature EER
angle + pen angle 5.6%
angle + pen angle velocity 5.6%
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Future work
Future work
• Extraction of new features - Pen lifting detection
• Stereo camera calibration - 3D information
• Connect with other device - pen with accelerometer
J´ozsef N´emeth Online Signature Feature Extraction
Online Signature
Feature Extraction
J´ozsef N´emeth
Contents Introduction The method
Recording Tracking Comparison Classification Results
Error measuring Results Future work
Acknowledgement
The presentation is supported by the European Union and co-funded by the European Social Fund.
Project title: ”Broadening the knowledge base and supporting the long term professional sustainability of the Research University Centre of Excellence at the University of Szeged by ensuring the rising generation of excellent scientists.”
Project number: T´AMOP-4.2.2/B-10/1-2010-0012
