Image Pattern Recognition
Download
Report
Transcript Image Pattern Recognition
Image Pattern Recognition
The identification of animal species through the
classification of hair patterns using image pattern
recognition: A case study of identifying cheetah prey.
Principal Investigator: Thamsanqa Moyo
Supervisors: Dr Greg Foster and Professor Shaun Bangay.
Presentation Outline
• Background (Why?)
• Objectives (What?)
• Approach to the Study (How?)
• Timeline (When?)
• Questions
Background to the Study
• Hair identification in Zoology and Forensics
• Manual image reference systems
Micrographs used in a Manual
System
D
V
Wildebeest (Gnu)
D
Jackal
V
V
D
Kudu
Background to the Study
• Recent Computer Image Identification Work
– With cheetah and zebra
Background to the Study
• How this project fits in:
– Builds on previous work
– Computer-based vs manual process
– Hair structure patterns vs skin patterns
Objectives of the Study
• To investigate:
– Image pattern recognition techniques
– Apply the techniques in hair pattern identification
• Resulting in a system that will:
– Report probable identities of hair patterns in images
– Be accurate enough to supplement manual efforts
Approach to the Study:
The Process
Patterns
Sensor
Image Manipulation
Feature
Feature
Classifier
System
Generation
Selection
Design
Evaluation
Artificial Intelligence
Figure Adapted from Theodoris et al (2003:6)
Approach to the Study:
Implementation
• Using ImageJ
– Image manipulation application
– Public domain application written in Java
• Plugins easily implemented in Java
Stage1: Sensor
• Producing input for the process
• Image manipulation based stage
• Images sourced from Zoology Department
Stage1: Sensor
(Grayscale)
Impala Patterns
(Binarized)
(Grayscale)
(Binarized)
Red Hartebeest Patterns
Stage 2: Feature Generation
• Identify features from patterns
– Hair patterns are feature vectors
• Define feature representation
• Larger than necessary number created
Stage 3:Feature Selection
• Selection of “best” features
• Considerations
– Computational complexity
– Capability of the classifier stage
• Produce training patterns (sets)
– Used in classifier design
Stage 4:Classifier design
• Place patterns into appropriate classes
• Linear and Non-Linear Classifiers
– E.g. neural network and perceptron
• Artificial Intelligence based stage
Stage 5: System Evaluation
• Assess Performance
– Use known and unknown patterns
– Compare with manual system
– Field trip
Timeline
Ta sks
W BS
Na m e
4 Sensor St age design
5 Feat ure Generat ion
6 Feat ure Selec t ion
7 Classifier Design
8 Sy st em Ev aluat ion
9 Possible field t rip t o t est sy st em
15 Projec t Hand in
Table 1: Extract of WBS found on Project Website
•238 days from 14 March
•Iterative Process
St a rt
Apr
Apr
May
Jun
Aug
Aug
Nov
15
22
13
22
12
19
07
Conclusion
• Background (why?)
– Useful to disciplines such as Zoology and Forensics
• Objectives (what?)
– Hair Pattern Recognition System
• Approach (How?)
– 5 stage approach(Graphics to AI)
• Timeline (When?)
– 238 days from 14 March
Questions
• Background (Why?)
• Objectives (What?)
• Approach (How?)
• Timeline (When?)