|
|
|
What is PT-UBIID? [^ top ^]
|
PT-UBIID is the first publicly available set of processing tools for the University of Bath Iris Image Database (UBIID - the free version containing 1000 eye images),
tools that can be used to generate test data sets (iris code databases), without wasting precious time. The toolbox is written in Matlab & ANSI C.
The second version of the toolbox (PT-UBIID-v.02) is available for DOWNLOAD (15.6 MB / 16,387,442 bytes).
Current functionalities:
- Iris Segmentation for UBIID: Circular Fuzzy Iris Segmentator 2 (CFIS2, [1], [3]-[5]);
- Iris Binary Encoding: Log-Gabor Encoder (LGE), Haar-Hilbert Encoders (HH1, HH2, [1]), Hilbert Encoder (GAITBE, [3], [5]);
- Iris Recognition Tests Simulator (12 single/multi-enrollment iris recognition tests [1]);
- Graphical display functions for generating figures and showing data statistics;
Main functions within the current version:
- 'tests.m' - run 12 iris recognition tests [1] on UBIID;
- 'cfis2.m' - the segmentator;
- 'hh1_encode_all.m', 'hh2_encode_all.m', 'lge_encode_all.m' - binary encoders;
- 'disp_fig_scores.m' / 'me_disp_fig_scores.m' - graphical display for the imposter and genuine score distributions (single/multi-enrollment);
- 'disp_score_statistics.m' / 'me_disp_score_statistics.m' - display data statistics for the current test in command window (single/multi-enrollment);
- 'ubiid_get_sim_scores.m' / 'me_ubiid_get_sim_scores.m' - compute similarity scores for the entire database (single/multi-enrollment);
PT-UBIID - Processing Toolbox for the University of Bath Iris Image Database
|
|
|
How can you use the toolbox? [^ top ^]
|
- If you aim to work with rectangular uint8 iris segments,
use CFIS2 (Circular Fuzzy Iris Segmentator, [1], [3]-[5]) to output all rectangular iris segments as variables or files.
You can also:
- analyse in unit8 domain the efect of pupil dilation, i.e. radial alignment problems (radial displacement),
- estimate a radial motion model, etc.,
- study ambiental light filtering, iris normalization, enhancement and occlusion removal.
- search for new binary encoding methods.
- search for a new feature-extraction technique to produce different kinds of iris codes in different feature vector spaces.
- Use LGE, HH1 and HH2 encoders to produce databases of binary iris codes of custom size.
Compare these codes directly or use them to train recognition memories or to search for new ways of expressing similarity (or distance) between the iris codes.
- Run 12 iris recognition tests (4'470'036 comparisons) and display their results.
- Build and run your own iris recognition tests.
|
|
|
INSTALLATION NOTES for PT-UBIID: [^ top ^]
|
- Compatibility Requirements for PT-UBIID-v.02.
- Matlab R13 or higher.
- Image Processing Toolbox (`imshow` is used).
- Wavelet Toolbox (`dwt2` is used)
- Signal Processing Toolbox (`hilbert` is used).
Even if you have none of these toolboxes, you still can use the segmentator CFIS2, the functions which compute and display the data statistics, Log-Gabor Encoder, and much of the code within`tests.m` file.
- Download the database.
- University Of Bath Iris Image Database - Free Version (1000 eye images) is available at the Department of Electronic & Electrical Engineering, University of Bath.
- During the tests undertaken to verify the results of our latest limbic boundary detector, we saw that the left and right eyes of the subject 17 are, in fact, interchanged. You should correct this problem when you have the database.
- Convert the eye images to a popular lossless graphic format (bmp, pgm, jpg).
The images within UBIID are *.j2c files. Convert them to a popular lossless graphic format (bmp, pgm, jpg) and preserve their original size of 960x1280. We recommend jpg.
Put them all in a directory 'X' (eliminate subdirectories like 'R' or 'L'). Be sure that the directory 'X' contains nothing else/more/less than the eye images extracted from UBIID at original dimensions (960x1280).
- Download and Extract 'PT-UBIID-v.02' archive in your Matlab work directory.
A directory 'PT-UBIID-v.02' is created together with another 3 sudirectories: 'DATABKP','OUT','TESTIMAGES'.
- Start Matlab and set 'PT-UBIID-v.02' as your current directory.
- Test the following commands:
- [ui,r] = cfis2('testimages\','',0,1); % this should work anyway
- [ui,r] = cfis2('testimages\','',1,1); % this will fail if you don't have the Image Processing Toolbox installed.
- [ui,r] = cfis2('testimages\','out\',0,1); % it will work even if you don't have the Image Processing Toolbox installed.
- tests % (see tests.m); it requires Image Processing Toolbox, Wavelet Toolbox at Signal Processing Toolbox.
- help cfis2 % use this help command to rapidly figure out who, what, where, why and why not.
- See the `tests.m` file in order to figure out what and where will be saved.
- Make the appropriate changes in the `Config` section of `tests.m` file and run the command: tests %
|
|
|
Examples of iris recognition results obtained using the toolbox [^ top ^]
|
Fig.1: Test T1 from [1]: Single-enrollment iris recognition test with iris codes on 4 Kilobits (512 Bytes). Statistics of all-to-all comparisons.
Fig.2: Test T2 from [1]: Single-enrollment iris recognition test with iris codes on 4 Kilobits (512 Bytes). Statistics of all-to-all comparisons.
Fig.3: Test T3 from [1]: Single-enrollment iris recognition test with iris codes on 4 Kilobits (512 Bytes). Statistics of all-to-all comparisons.
Fig.4: Test T7 from [1]: Multi-enrollment iris recognition test with iris codes on 4 Kilobits (512 Bytes). Each enrolled identity is defined by 10 gallery templates. Statistics of Mean-Deviation Similarity Scores obtained by comparing all candidate templates to all enrolled identities.
Fig.5: Test T8 from [1]: Multi-enrollment iris recognition test with iris codes on 4 Kilobits (512 Bytes). Each enrolled identity is defined by 10 gallery templates. Statistics of Mean-Deviation Similarity Scores obtained by comparing all candidate templates to all enrolled identities.
Fig.6: Test T9 from [1]: Multi-enrollment iris recognition test with iris codes on 4 Kilobits (512 Bytes). Each enrolled identity is defined by 10 gallery templates. Statistics of Mean-Deviation Similarity Scores obtained by comparing all candidate templates to all enrolled identities.
|
|
|
The motivations behind the realease of PT-UBIID: [^ top ^]
|
- Currently, Iris Recognition is not quite accessible for individual researchers. And this is
mainly because it requires a sum of strong theoretical and practical skills in Digital Image Processing, Digital Signal Processing,
Artificial Intelligence, Statistics, and at last, but not the least, in Software Engineering.
Hence, PT-UBIID will make Iris Recognition easier for individual researchers.
- In our view, the most difficult part of Iris Recognition is Iris Segmentation,
i.e the stage in which the actual iris segment captured within an eye image is mapped to (extracted as) a rectangular iris segment
(usually, a uint8 matrix whose height and width are - for obvious reasons - multiples of 8 and 64, respectively).
- Programming Iris Recognition is nothing more than replicating pieces of human
knowledge, perception and reasoning in a computable manner.
We all know (or assume) that iris texture is visually recognizable by humans.
Hence, it is obvious that, all in all, implementing iris recognition is a matter of expressing a specific human behaviour
and a particular hypostasis of human intelligence by using computational means. But in a scenario in which segmentation
is wrong, the extracted iris segments will be neither visually nor computationally recognizable.
- We are pretty confident that releasing the first publicly available package of processing tools
for University of Bath Iris Image Database (UBIID) will boost the general interest for iris recognition in the
computational/artificial intelligence research community, mainly due to the fact that generating test datasets (iris code databases)
will become easier and less time-consuming than today.
|
|
|
References: [^ top ^]
|
- 2010: Nicolaie Popescu-Bodorin, Valentina E. Balas,
Comparing Haar-Hilbert and Log-Gabor based iris encoders on Bath Iris Image Database
,
4th International Workshop on Soft Computing Applications,
July 15-17, 2010, Arad, ROMANIA.
- 2010: Nicolaie Popescu-Bodorin, Valentina E. Balas,
AI Challenges in Iris Recognition. Processing Tools for Bath Iris Image Database
,
11th Int. Conf. on Automation & Information (ICAI'10),
June 13-15, 2010, Iasi, ROMANIA.
- 2010: N. Popescu-Bodorin,
'Fragile Bits' vs. Multi-Enrollment - A Case Study of Iris Recognition on Bath University Iris Database
,
ROMAI Journal, Nr.5, vol. 2/2009, pp. 127-144, ISSN: 1841-5512 (print) / 2065-7714 (online), February 2010.
- 2009: N. Popescu-Bodorin,
A Fuzzy View on k-Means Based Signal Quantization with Application in Iris Segmentation
(short ppt overview here)
,
17th Telecommunications Forum - TELFOR 2009
(IEEE Conference #15805),
IEEE Serbia & Montenegro COM Chapter and Section,
University of Belgrade,
Telecommunications Society - Belgrade,
Telekom Srbija ad,
PTT Communications "SRBIJA",
24-26 November 2009, Belgrade, SERBIA;
- 2009: Nicolaie Popescu-Bodorin,
Exploring New Directions in Iris Recognition
,
11th
International Symposium on Symbolic and Numeric Algorithms for Scientific Computing,
Conference Publishing Services - IEEE Computer Society,
pp. 384-391, DOI: 10.1109/SYNASC.2009.45.
- 2009: Nicolaie Popescu-Bodorin,
Gabor Analytic Iris Texture Binary Encoder
,
Proceedings of the 4th Annual South East European Doctoral Student Conference,
vol. 1, pp. 505-513, ISBN 978-960-9416-00-9, 978-960-9416-02-3,
ISSN 1791-3578, South-East European Research Centre (SEERC),
July 2009.
|
|
|
|
|
|
|
Page maintained by Nicolaie Popescu-Bodorin,
Contact (e-mail): bodorin # ieee. org
Last update: February 09, 2013
|
