Title: Face Recognition: Evaluation Report for FaceIt Identification and Surveillance
1Face Recognition Evaluation Report for FaceIt
Identification and Surveillance
Jingu Heo The department of Electrical Computer
Engineering The University of Tennessee,
Knoxville
2FaceIt Identification Template
2. Select a Subject
1. Create Gallery
3.Matching
3Overall Test with FERET and IRIS DB
4Detailed Tests- Expression Illumination
5Detailed Tests Age
6Detailed Tests -Pose
7Detailed Tests Head Size
8095
90110
100120
125150
150180
1012
2024
3036
4048
5065
6072
7085
256384 pixels Actual Size
8Detailed Tests Head Size
Auto aligning Fails
Auto aligning -gt At least 20 pixels between the
eyes
7085
8095
150180
125150
100120
90110
1012
6072
3036
2024
5065
4048
9Detailed Tests Head Size
Head size Test (Gallery(ba)-200, Subject(Cropped
ba)-200)
10Summary of FaceIt Identification Results
11FaceIt Surveillance Template
1. Create Gallery
2. Capture faces from Video
3. Matching
12FaceIt Surveillance Test
- Overall Test - IRIS, FERET
- 758 captured faces / 13 individuals
- Distance 23ft from the camera
- Detailed Test IRIS, FERET
- Facial Variations (Small or Large)
- Database size(Small or Large)
- Different Illumination
- Distance(Low resolution Image)
13Overall Test (Small DB,Large Variations,
HighFrontal Lighting)
14Example of Captured images
15Detailed Tests DB size and Variations
16Detailed Tests Lighting and Distance
17Conclusion
- Some faces which has more distinctive features
are more easily recognized than others. - Mixed variations(AgePose Illumination
Expressions) affects more than single variation - Face size is not big problem as long as the
features are clear. - Small DB, small variation, high and additional
frontal lighting and frontal faces give us the
best reliability.