Download presentation
Presentation is loading. Please wait.
Published byTamsyn Owen Modified over 8 years ago
1
Iris-based Authentication System Daniel Schonberg and Darko Kirovski, “Iris Compression for Cryptographically Secure Person Identification”, in Proceedings of the Data Compression Conference 2004 (DCC’04) Multimedia Security
2
2 Outline Block Diagram of the EyeCert System Related Work Iris Detection Transformation Feature Extraction Data Compression Performance
3
3 Block Diagram of the EyeCert System
4
4 Related Work J. Daugman. Recognizing Persons by Their Iris Patterns. Biometrics: Personal Identification in Networked Society, Kluwer Academic Publishers, 1999. L. Ma, Y. Wang, and T. Tan. Iris Recognition Based on Multichannel Gabor Filtering. Asian Conference on Computer Vision, pp.23–25, 2002. C.-l. Tisse, et al. Person identification technique using human iris recognition. Journal of System Research, vol.4, pp.67–75, 2003. (Hough transform) L. Ma, et al. Iris recognition using circular symmetric filters. International Conference on Pattern recognition, vol.2, pp.414– 417, 2002.
5
5 Iris Detection Detect the pupil Detect the outer edge of the iris Remove the noise (skin, eyelid, eyelashes, etc.)
6
6 Transformation The Fourier Mellin Transform (FMT) Be related to the Fourier transform
7
7 Transformation (cont.) 2D Three Steps: –Interpolation (bicubic) –Scaling –Fourier Transform
8
8 Transformation (cont.) 256 angles
9
9 Feature Extraction Similarity Select the proper set of features
10
10 Feature Extraction (cont.) Select the top 170 data points of the phase of mFMT to create personal feature vector
11
11 Data Compression There are THREE data-sets that need to be compressed: –The value of the data in the feature vector –The location of the selected mFMT coefficients –The decision threshold
12
12 Data Compression (cont.) The feature vector –-π ≦ θ ≦ π –5 bits per sample –Total 170 * 5 = 850 bits The location of the coefficients –Using off-the-shelf point compression algorithm which uses delta-modulation and combinatorial optimization –6 bits per location –Total 170 * 6 = 1020 bits The decision threshold –10 bits per threshold 1880 bits per user
13
13 Performance System I – train on all 108 persons within the CASIA database System II – train on a random selected 90 users
14
14 Performance (cont.)
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.