BEST PRACTICES FOR FACIAL RECOGNITION USING MOBILE DEVICES
The purpose of this study was to determine the best practices and characteristics for capturing a quality image with a mobile device that meets standards for facial recognition software. The focus of our study on image quality revolved around lighting, subject pose, and camera properties. Our analysis of image quality was based off of ISO standards for facial recognition.
Chris Nagelbach, James Sternberg, Cody Maus, William Payne, Michael Brockly, Stephen Elliot
Overview
Distance Angle Type
Lighting 1.5 m, depth of field 10
cm
45 degrees left/right center, 35 degrees
above center
Dual Lighting, 800-900 Lux
Pose/Person
Eye width should be 25% of picture width and located at 60% of
total image height
+/- 5 degree from center
Neutral emotion with naturally open eyes, 0
motion blur
Camera Between 0.5 to 1 m +/- 5 degrees
above/below eye level Back facing camera
Background N/A N/A Illuminated 18% grey
Best Practices
Camera Distance From Face Eye Separation Compliant Failure Rate
0.5M 436.53 OK 0%
1.0M 182.48 OK 0%
1.5M 129.7 OK 0%
2.0M 104.52 Fail 100%
Lighting Type Percent Facial Brightness Compliant Failure
Rate Overhead Fluorescent
Lighting 50.65 OK 7%
2 Studio Lamps at Eye Level 45˚ From Center 60.86 OK 0%
2 Studio Lamps 35˚ Above Eye level 45˚ from center 61.13 OK 0%
Percent Background Gray Compliant Failure
Rate Percent Facial
Brightness Compliant Failure Rate
Eye Separation Compliant Failure
Rate Rear facing camera with white
background 12.65673772 OK 0% 47.3 OK 0% 477.9623327 OK 0%
Front facing camera with white background 8.808231499 OK 0% 34.7 Fail 30% 122.5834791 Fail 10%
Rear facing camera with 18% grey background 42.66904296 OK 0% 62.9 OK 0% 607.0340979 OK 0%
Front facing camera with 18% grey background 39.84486436 OK 0% 57.6 Ok 0% 121.4669226 Fail 20%
Conclusions
Camera Distance From Face Lighting
Camera and Background
Ideal Experiment Setup
Based on our research and experimentations we have determined the ideal best practices listed in the best practices table above. We came to this conclusion by taking 10 pictures for each of the desired variables on an iPhone 4s. The pictures were processed against the ISO_FRONTAL_Best_Practices standard. The output from this analysis gave us statistical data to determine the best practices for a mobile device testing environment.
D ’ Amato, D. (n.d.). Best practices for taking face photographs and face image quality metrics. (2006). NIST Biometric Quality Workshop