May 14, 2008 الله أكبر و لله الحمد 1 Automated Identification Systems Hany Ammar Lane Dept. of Computer Science & Electrical Engineering The 2 nd International.
Post on 16-Jan-2016
212 Views
Preview:
Transcript
May 14, 2008 الحمد لله و أكبر الله 1
Automated Identification Systems Hany Ammar
Lane Dept. of Computer Science & Electrical Engineering
The 2nd International Conference on Computer andCommunication Engineering (ICCCE08), KL, Malaysia
A key-Note Presentation on
الرحيم الرحمن الله بسمالله رسول على السالم و الصالة و لله الحمد
April 9, 2008 الحمد لله و أكبر الله 2
OutlineOutlineAutomated Identification Systems
The Center for Identification Technology Research (CITeR)
Examples of Automated Identification Systems Projects
Automated Dental Identification Systems (ADIS) Research Team Funding Agencies Overview of ADIS and the ADIS Architecture Record Pre-processing Dental Image Retrieval Matching
Summary
April 9, 2008 الحمد لله و أكبر الله 3
Automated Identification Systems
Automated identification of a person based on his/her physiological or behavioral characteristics Termed as “Biometrics”Identification
Fingerprint Hand Geometry Signature
Dental Features Iris Voice
April 9, 2008 الحمد لله و أكبر الله 4
Automated Identification Systems
APPLICATIONS INCLUDE HIGH SECURITY APPLICATIONS: financial
services, health care, law enforcement, Government applications, travel and immigration, and E-commerce
FORENSIC IDENTIFICATION: help solve legal cases and public issues which include bank robberies, homicides, kidnapping cases, and identifying victims of mass disasters (Post Mortem identification)
April 9, 2008 الحمد لله و أكبر الله 5
Automated Identification Systems
Forensic Post-Mortem (PM) Identification Methods include:
- Visual- Fingerprints- DNA- Dental
Dental features- Used to identify 75% of Tsunami victims in Thailand,
and 20% of 9/11 victims identified in the 1st year compared to only 0.5% identified using DNA
- Resist early decay of body tissues.- Withstand severe conditions in mass disasters.- Unique (Identification can sometimes be made from
one tooth).
April 9, 2008 الحمد لله و أكبر الله 6
Automated Identification Systems
Example systems Automated Dental Identification System
ADIS
ADIS
Digital Image Rep
Mrs. X PM Record: - NCIC codes- Dental Radiographs
Short Match List
Forensic Scientist
April 9, 2008 الحمد لله و أكبر الله 7
Automated Identification Systems
Example systems Automated Ear Identification System
AEISVideo
Sequence
Ear Segmentation and Localization
Image Enhancement
2-D and 3-D Feature Extraction
IdentificationEnrollment
Decision
Data-base
Currently being developedWVU-UM
April 9, 2008 الحمد لله و أكبر الله 8
Automated Identification Systems
Biometrics Lab at WVU – Face Video data acquisition system Collected a
Database of 500Subjects
April 9, 2008 الحمد لله و أكبر الله 9
NSF Center at WVUCITeR
The US National Science Foundation Center for Identification Technology Research (CITeR)Industry/University Cooperative Research Center (I/UCRC)West Virginia University is the lead institution http://www.citer.wvu.edu/
April 9, 2008 الحمد لله و أكبر الله 10
OutlineOutlineAutomated Identification Systems
The Center for Identification Technology Research (CITeR)
Examples of Automated Identification Systems Projects
Automated Dental Identification Systems (ADIS) Research Team Funding Agencies Overview and the ADIS Architecture Record Pre-processing Dental Image Retrieval Matching
Summary
April 9, 2008 الحمد لله و أكبر الله 11
ADIS Project Research Team
Prof. Hany Ammar, Dr. Gamal Fahmy, Dr. Robert Howell, Dentist,
Ph.D. Students: Ayman Abaza, Diaa Nassar, Eyad Haj-Said,
MS Students: Mubasher, Zainab Millwallah, Usman Qureishi, Faisal Chaudhry, Mythili, and Satya Checkuri, Ali Bahoo
Prof. Anil Jain,Ph.D. Student: Hong Chen
Prof. Mohammad AbdelMottaleb,Ph.D. Students: Omaima Nomair, Mohammad Mahoor, Jindan,
April 9, 2008 الحمد لله و أكبر الله 12
Support
$1.5M over 5 years- This research is supported in part by the U.S. National Science Foundation (Digital Government Program) under Award number EIA-0131079 to West Virginia University, - The research is also supported under Award number 2001-RC-CX-K013 from the Office of Justice Programs,National Institute of Justice, U.S. Department of Justice. Points of view in this document are those of the authorsand do not necessarily represent position of theU.S. Department of Justice.- The research is conducted in Collaboration with The Criminal Justice Information Services Division (CJIS) of the US Federal Bureau of Investigation
April 9, 2008 الحمد لله و أكبر الله 13
Forensic Odontologist Compares PM Records with AM records based on:
- Dental Work (e.g. Fillings, Restorations ...)
- Inherent Dental Characteristics (Crown Morphology, Root Morphology, Spacing …)
- Very Time Consuming Process
OverviewOverviewDental Identification
Identification of the victims of 9/11
- 20% of the 973 identified in the first year
- Identification of 2,749 took around 40 months.
April 9, 2008 الحمد لله و أكبر الله 14
Source: The Bureau of Legal Dentistry (BOLD) - http://www.boldlab.org [2000]
OverviewOverviewDental Identification is a challenging problem
AM
PM
April 9, 2008 الحمد لله و أكبر الله 15
ArchitectureOverviewOverview
April 9, 2008 الحمد لله و أكبر الله 16
ADIS OutlineADIS OutlineOverview
Record Pre-processing
Dental Image Retrieval
Matching
Conclusion & Future Work
Comments & Questions
April 9, 2008 الحمد لله و أكبر الله 17
Record Pre-processingRecord Pre-processing1- Record Cropping:
global segmentation of dental films from their corresponding records.
The objective:
to automate the process of cropping a composite digitized dental record into its constituent films
Reference Record - 16
Subject Record
April 9, 2008 الحمد لله و أكبر الله 18
Dental Record
Pre-Processing
Cropping based on
Arch-Detection
Round Right
Cropping based on
Factor Analysis
Corner-type
Classification
Background
Extraction
Post-Processing
Dental Films
Record CroppingRecord CroppingApproach
April 9, 2008 الحمد لله و أكبر الله 19Under-segmented
Record CroppingRecord CroppingExperimental Results
April 9, 2008 الحمد لله و أكبر الله 20
By calculating “”,
“” found to range between 0.49 - 0.91,
“” was used to identify the Under Cropped Segments.
Record cropping time ranges 15-40 sec.
},min{ hw
wh
Record CroppingRecord CroppingExperimental Results
Perfectly
Cropped
74%
Under Cropped
24%
Error
2%
Perfectly Cropped
Under Cropped
Error
Randomly selected test sample of 100 dental records (images) from the CJIS ADIS database, the total film count in the test set is 722.
April 9, 2008 الحمد لله و أكبر الله 21
Record Pre-processingRecord Pre-processing3- Film Type Detection:
dental films classification into bitewing, periapical, or panoramic.
The objective:
to automate the process of dental film type detection.
bitewingperiapical
panoramic
April 9, 2008 الحمد لله و أكبر الله 22
Record Pre-Record Pre-processingprocessing4- Teeth Segmentation:
Teeth segmentation from dental radiographic films.
The objective: to automate the process of local segmentation of each tooth. teeth isolation into a rectangular box
April 9, 2008 الحمد لله و أكبر الله 23
Record Pre-processingRecord Pre-processing5-Tooth Contour Extraction:
another level of segmentation, to extract the contour of the tooth.
The objective:
to extract an accurate smooth representative tooth contour,
- Representative smooth contour.
- Time / tooth = fraction of the second).
April 9, 2008 الحمد لله و أكبر الله 24
Record Pre-Record Pre-processingprocessingExperimental Result
Records
Correct or partially correct contour extraction (%)
Errors(%)
Average time (s)
Perfect contour
(P)
Perfect crown (PC)
Partially correct
(C)
Errors(E)
10 AM 56.2 14.05 16.75 13.0 0.15
10 PM 60.0 11.61 14.83 13.56 0.17
ALL 58.10 12.83 15.79 13.28 0.16
The snake-based algorithm on the same platform takes about 5 sec compared to 0.16 sec.
For a test set of 20 records, involving ~340 teeth
April 9, 2008 الحمد لله و أكبر الله 25
Record Pre-processingRecord Pre-processing6-Teeth Labeling: automatic classification of teeth into incisors, canines, premolars and molars as part of creating a dental chart.
The objective: - to accurately classify and label teeth,
- to accommodate a missing segment.
RX7RX6 RX5
RX4
RD7RD6 RD5
RD4
7M 5
P
April 9, 2008 الحمد لله و أكبر الله 26
An adult has 32 permanent teeth
(8 Incisors, 4 Canines, 8 Premolars and 12 Molars).
Each tooth has a specific structure and position in the mouth.
Dental Atlas for the left half of the upper jaw.
Record Pre-processingRecord Pre-processingDental Atlas
American Medical Association, http://www:medem:com
April 9, 2008 الحمد لله و أكبر الله 27
- Teeth Classification: added the film type, designed a technique based on Linear DiscriminantAnalysis (FisherTeeth).
- Extended the validation stage for the presence of missing tooth.
Teeth Labeling Approach – Eigen Teeth Labeling Approach – Eigen Teeth Teeth
Record Pre-Record Pre-processingprocessing
April 9, 2008 الحمد لله و أكبر الله 28
Experimental Results of teeth labeling
Based on the dataset used in the literature,
(50 bitewing films involving about 400 teeth).
Method Molars Average
Premolars Average
Labeling Time
Complex Signature 89.6% 90.95% 21.3 msec
Centroid Distance 90.55% 87.85% 21.3 msec
Eigen Teeth 91.67% 92.86% 11.5 sec
Record Pre-processingRecord Pre-processing
April 9, 2008 الحمد لله و أكبر الله 29
ADIS OutlineADIS OutlineOverview
Record Pre-processing
Dental Image Retrieval
Matching
Conclusion & Future Work
Comments & Questions
April 9, 2008 الحمد لله و أكبر الله 30
Dental Image Retrieval4- Potential Matches Search: searching the dental database in a fast way to find a candidate list.
The objective: - to accomplish a relatively short candidate list, with a high probability of having the correct match reference.
This objective directly targets the scalability of ADIS system.
Candidate List
Digital Image Repositories
April 9, 2008 الحمد لله و أكبر الله 31
Potential Match Potential Match SearchSearchChallenges
Multiple RepresentationOf the same tooth (RX6)
Reference Record
Subject Record
April 9, 2008 الحمد لله و أكبر الله 32
Potential Match Potential Match SearchSearchProposed Approaches
Archiving
Retrieval /Matching
Preprocessing Stage
LabeledTeeth
Segments
View Normalization
Appearance-based
features
CandidateList
Subject Preprocessing
Reference Preprocessing
Teeth Contour Extraction
Shape-based
features
Digital Image Repositories
1- Appearance-based, namely Eigen images.
low computational-cost features;
Limitation: need geometric and gray-scale normalization.
2- shape –based namely moment invariant and edge orientation histogram Limitation: need accurate teeth contour.
April 9, 2008 الحمد لله و أكبر الله 33
Potential Match Potential Match SearchSearchExperimental Result (Comparison between appearance and shape
based)
Minimum fusion, better for shape-based.
The appearance-based, better for short candidate list.
The edge direction histogram achieves the same performance for slightly longer candidate list. 0 5 10 15 20 25 30 35 40 45
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Candidate List Size
Hit
Rat
e
Best Min Fusion CMC curves
OneClassFourClassSmoothHist
April 9, 2008 الحمد لله و أكبر الله 34
ADIS OutlineADIS OutlineOverview
Record Pre-processing
Dental Image Retrieval
Matching
Conclusion & Future Work
Comments & Questions
April 9, 2008 الحمد لله و أكبر الله 35
Image Comparison Image Comparison ComponentComponent
April 9, 2008 الحمد لله و أكبر الله 36
Image Comparison Image Comparison ComponentComponent
Teeth Alignment: is to align each corresponding pair, in other word to find the transformation matrix that best align the reference and subject segments.
The objective: is to achieve an accurate aligned segments in few seconds, so as to allow for a faster Image Comparison Component.
Teeth Alignment
April 9, 2008 الحمد لله و أكبر الله 37
A Hierarchical fusion scheme: Tooth-level fusion Case-level fusionA Ranking Scheme to Sort the Match List
Micro and Macro Decision-Making (The Strategy)
Image Comparison ComponentImage Comparison Component
April 9, 2008 الحمد لله و أكبر الله 38
Results
Image Comparison ComponentImage Comparison Component
April 9, 2008 الحمد لله و أكبر الله 39
OutlineOutlineAutomated Identification Systems
Example Research Projects
Automated Dental Identification Systems (ADIS)
Research Team Funding Agencies The ADIS Architecture Record Pre-processing Dental Image Retrieval Matching
Summary
April 9, 2008 الحمد لله و أكبر الله 40
SummarySummary
• Automated Identification Systems are needed in many applications in the yearsto come
• They Pose many challenging problems
April 9, 2008 الحمد لله و أكبر الله 41
SummarySummary
Timeliness Performance Teeth labeling and alignment are
time consuming processesQuality of radiographs are very critical for ADIS
Poor quality can affect the segmentation accuracy significantly
Matching efficiency can also be affected by poor quality radiograph
ADIS challenges
top related