User Verification System by William Baker, Arthur Evans, Lisa Jordan, Saurabh Pethe Client Dr.Cha
Aim: To improve confidence level by hybridizing multiple biometrics such as Face, Finger print, Handwriting, Hand geometry, Iris and Voice. Confidence Level: Percentage of correct answers, valid user accepted and invalid user rejected. To reduce false positive and false negative errors : - valid user rejected - invalid user accepted
Types of Biometrics decided for this project experiment: Face Handwriting Voice Finger print
biomouse Fingerprint scanner Digital Camera LCD Pen tablet Microphone Multi-modality Biometric Authentication Embeded & Hybrid User Verification system System that requires user verification
Hand Writing features: Width Height Drag count Total stroke time Total stroke distance Stroke direction sequence string Acceleration Tools used: LCD Pen Tablet for data collection Java application for feature extraction
Each person writes differently.
Face Recognition: Photos collected have to be properly sized and also be gray scale. Eigen face technology is used to calculate the mean face/value Recognition is done using Nearest Neighbor method. Tools Used: Digital Camera for data collection Mathworks’ Matlab for training and recognition
Each person has different faces.
? Query Face DB Face Recognition System
width, length User 1 User 2 User1 s1 = ( 12, 16 ) User1 s2 = ( 11, 20 ) User2 s1 = ( 9, 8 ) User2 s2 = ( 10, 7 ) Truthfeatures Measurements
slant width user1 user2 = user1 ? Nearest Neighbor Classifier too slow for users to wait for the output.
Data Acquisition Feature Extraction Training an ANN Classification System HandwritingDone -- FaceDone** VoiceDone--- Finger print ---- Modality Steps Project Status ** - Eigen face and nearest neighbor methods used.
Advantages: Higher accuracy of determining an individual Reliable by having multiple recognition techniques or biometrics Increased security in companies Reduced amount of time to identify a suspect or criminal for law enforcement Difficult to challenge the system by forging names and mimicking voices making it virtually impossible to pass as someone else Possible use in a court of law to prove criminal cases Low maintenance software
Future Plan: Handwriting training and classification. Voice feature extraction methods Finger print data collection
Demonstration Handwriting Face Recognition
Sub-classing with Java