Project Sunbeam
Post on 23-Feb-2016
50 Views
Preview:
DESCRIPTION
Transcript
Project SunbeamMobile Iris Recognition
Normalized Original
Pupil Detection
Limbic Detection
Segmentation Result
Unwrapped
Project Goals and TimelineProblem Statement: To-date there is no mainstream mobile platform for Iris Biometrics that does not require either dedicated hardware or tethering to a computer.
As of March 2012¹: 50.4% of U.S. adults own a smartphone. 24.4% Android. 16.1% iOS.
Project Sunbeam aims to create “One-to-Many” Iris Recognition Platform for Mobile Devices such as Apple iOS and Google Android operating systems.
Platform Segmented into Two Main Components:
Library (Framework) Focus on the algorithms designed to segment and produce identifiable iris signatures (Iris Bitcode).
Mobile Application (Front-End) Focus on Mobile Application Implementation and Mobile Platform Limitations. Mobile Application implements
Framework.
Project Goals • Pupil Detection • Going Forward • More Information
¹Neilsen Ratings: http://blog.nielsen.com/nielsenwire/?p=31688
Project’s current focus is on finishing the framework to a complete working state.
Recently finished Entire Process to Produce Iris Signature (Iris Bitcode) from Normalized Image. Framework approx. 95% functional.
Project Goals and Timeline
2048-Byte Iris Bitcode
(1/8 of the Entire Bitcode)
Segmented Iris(Polar Coordinates)
Segmented Iris (Cartesian Coordinates)
Normalized Image Segmented Image
Project Goals • Pupil Detection • Going Forward • More Information
Pupil DetectionOne of the first tasks for the Framework was to segment the Pupil from the Iris.
How? Because the pupil has the lowest intensity (darkest color) we expect that we can look at the entire image and determine which pixels are apart of the pupil.
First we look at each pixel in the image and find the average intensity over the entire image. We then calculate an intensity threshold.
Project Goals • Pupil Detection • Going Forward • More Information
Pupil DetectionHow? Using the Intensity Threshold we now check every pixel in the image against the threshold. If a single pixel’s intensity is below the
threshold, we conclude that it is part of the pupil.
Project Goals • Pupil Detection • Going Forward • More Information
Pupil DetectionProblem?Because eye lashes, makeup, and other portions of the face can also
have a low intensity, they can be mistaken as part of the pupil, misaligning the pupil location.
Solution We found through empirical testing that the initial guess, although misaligned, is within close proximity of the actual pupil.
We now look at only the pixels within a square of 25% width and height around the initial guess.
This now allows us to accurately detect the pupil’s location and size.
Project Goals • Pupil Detection • Going Forward • More Information
Pupil DetectionProject Goals • Pupil Detection • Going Forward • More Information
Pupil DetectionProject Goals • Pupil Detection • Going Forward • More Information
Going ForwardShort Term Complete Framework to a 100% working solution.(This Semester) Involve more test cases and adjust/improve algorithms as necessary.
Empirical Iris Bitcode matching testing and statistics.
Long Term Start of Mobile Application.(Next Semester) Iris characterization and matching optimization.
Final Result Fully-working proof of concept Mobile Iris Recognition platform.
Project Goals • Pupil Detection • Going Forward • More Information
More InformationWhat we covered…
The need for a Mobile Iris Recognition Platform.No current solution for iOS or Android devices.No current mobile solution that does not require additional hardware.
Overview of the Pupil Segmentation Process.Determining the boundaries and size of the pupil.
Project Goals • Pupil Detection • Going Forward • More Information
More InformationFor more information: http://sunbeam.tech.mtu.edu.
Project website will be periodically updated with progress updates, newer images, and code segments.
Matthew Ellison mmelliso@mtu.eduKyle Johnson kydjohns@mtu.edu
Advisor: Professor Hembroff
Project Goals • Pupil Detection • Going Forward • More Information
top related