Introduction to Biometrics Dr. Bhavani Thuraisingham The University of Texas at Dallas Lecture #11 Biometric Technologies: Iris Scan September 28, 2005 Outline Introduction Components Iris Scan Process Template generation and matching Market and Deployment Strengths and Weaknesses Research Directions Conclusions Appendix: Updated information on project References Course Text Book, Chapter 6 http://www.biometricsinfo.org//irisrecognition.htm Introduction Iris scan biometrics employs the unique characteristics and features of the human iris in order to verify the identity of an individual. The iris is the area of the eye where the pigmented or colored circle, usually brown or blue, rings the dark pupil of the eye. The iris-scan process begins with a photograph. A specialized camera, typically very close to the subject, no more than three feet, uses an infrared imager to illuminate the eye and capture a very high-resolution photograph. This process takes only one to two seconds and provides the details of the iris that are mapped, recorded and stored for future matching/verification. Introduction (Continued) The inner edge of the iris is located by an iris-scan algorithm which maps the iris’ distinct patterns and characteristics. An algorithm is a series of directives that tell a biometric system how to interpret a specific problem. Algorithms have a number of steps and are used by the biometric system to determine if a biometric sample and record is a match. Iris’ are composed before birth and, except in the event of an injury to the eyeball, remain unchanged throughout an individual’s lifetime. Introduction (Continued) Iris patterns are extremely complex, carry a large amount of information and have over 200 unique spots. The fact that an individual’s right and left eyes are different and that patterns are easy to capture, establishes iris-scan technology as one of the biometrics that is very resistant to false matching and fraud. The false acceptance rate for iris recognition systems is 1 in 1.2 million, statistically better than the average fingerprint recognition system. Introduction (Continued) Iris-scan technology has been piloted in ATM environments in England, the US, Japan and Germany since as early as 1997. In these pilots the customer’s iris data became the verification tool for access to the bank account, thereby eliminating the need for the customer to enter a PIN number or password. When the customer presented their eyeball to the ATM machine and the identity verification was positive, access was allowed to the bank account. These applications were very successful and eliminated the concern over forgotten or stolen passwords and received tremendously high customer approval ratings. Introduction (Concluded) Airports have begun to use iris-scanning for such diverse functions as employee identification/verification for movement through secure areas Allowing registered frequent airline passengers a system that enables fast and easy identity verification in order to expedite their path through passport control. Other applications include monitoring prison transfers and releases, as well as projects designed to authenticate on-line purchasing, on-line banking, on-line voting and on-line stock trading. Components of the Iris Scan System Front-end acquisition hardware with central processing software Software components: Image processing and matching engines, Proprietary database Web-enabled iris scan applications are being developed Integration with middleware systems Process Iris-Scan: How it Works: Dr. John Daugman's work in iris recognition form the basis of this information. Information and images found on his website, http://www.cl.cam.ac.uk/users/jgd1000, are presented below. Iris recognition leverages the unique features of the human iris to perform identification and, in certain cases, verification. IrisCode (TradeMark) Misidentification rate Process: The Iris Iris recognition is based on visible (via regular and/or infrared light) qualities of the iris. A primary visible characteristic is the trabecular meshwork (permanently formed by the 8th month of gestation), a tissue which gives the appearance of dividing the iris in a radial fashion. Other visible characteristics include rings, furrows, freckles, and the corona Process: IrisCode (TradeMark) Iris recognition technology converts these visible characteristics as a phase sequence into a 512 byte IrisCode(tm), a template stored for future identification attempts. From the iris' 11mm diameter, Dr. Daugman's algorithms provide 3.4 bits of data per square mm. This density of information is such that each iris has ‘ 266 'degrees of freedom', as opposed to 13-60 for traditional biometric technologies. After allowing for the algorithm's correlative functions and for characteristics inherent to most human eyes, Dr. Daugman concludes that 173 "independent binary degrees-of-freedom" can be extracted from his algorithm - an exceptionally large number for a biometric. A key differentiator of iris-scan technology is the fact that 512 byte templates are generated for every iris, which facilitates match speed (capable of matching over 500,000 templates per second) Process: Iris Acquisition The first step is location of the iris by a dedicated camera no more than 3 feet from the eye. After the camera situates the eye, the algorithm narrows in from the right and left of the iris to locate its outer edge. This horizontal approach accounts for obstruction caused by the eyelids. It simultaneously locates the inner edge of the iris (at the pupil), excluding the lower 90° because of inherent moisture and lighting issues. Process: Iris Scan issues Iris-scan technology requires reasonably controlled and cooperative user interaction - the enrollee must hold still in a certain spot, even if only momentarily. In applications whose user interaction is frequent (e.g. employee physical access), the technology grows easier to use. Applications in which user interaction is infrequent (e.g. national ID) may encounter ease-of-use issues. Over time, with improved acquisition devices, this issue should grow less problematic. Process: Iris Scan issues (Concluded) The accuracy claims associated with iris-scan technology may overstate the real-world efficacy of the technology. Because the claimed equal error rates are derived from assessment and matching of ideal iris images (unlike those acquired in the field), actual results may not live up to the projections provided by leading suppliers of the technology. Since iris technology is designed to be an identification technology, fallback procedures may not be as fully developed as in a verification deployment (users accustomed to identification may not carry necessary ID, for example). Image Acquisition Kiosk-based systems - User stands 2-3 feet from camera positioned at the height of the user’s eye Physical access devices - Small camera mounted behind a mirror acquires the image. User locates his eye on the mirror Desktop cameras - 18 inches from device; PC/Workstation-based Inage Processing Process of mapping iris is the same for any acquisition device After camera locates the eye, an algorithms narrows in from the right and left of the eye to find the iris’s outer edge Iris scan algorithm locates the inner edge of the iris at the pupil - Challenging for very dark eyes Once the parameters of the iris have been defined a black and white image of the iris is used for feature extraction Distinctive Characteristics Primary visible characteristic is Trabecular Meshwork, a tissue that gives the appearance of dividing the iris in a radial fashion Others include: Rings, Furrows, Freckles, and Corona Maps segments of the iris into hundreds of independent vectors Characteristics derived from the iris features are the orientation and the spatial frequency of distinctive areas along with the position of the areas Not all of the iris is used Template Creation/Generation Vectors located by the iris scan algorithm are used to form enrollment and match templates Templates are generated in hexadecimal format Between one and four Iris images are needed for enrollment and template generation Template Matching Usually identification is performed more than verification - Template is matched against the ones in the database to identify the person Verification is performed infrequently - Matches a person’s iris template against the one stored for him/her in the database Application Market and Deployment Iris-scan technology has traditionally been deployed in high- security employee-facing physical access implementations Iridian - the technology’s primary developer - is dedicated to moving the technology to the desktop, and has had some success in small-scale logical access deployments. The most prominent recent deployments of iris-scan technology have been passenger authentication programs at airports in the U.S., U.K., Amsterdam, and Iceland The technology is also used in corrections applications in the U.S. to identify inmates. A number of developing countries are considering iris-scan technology for national ID It is believed that the largest deployed Iridian database spans under 100,000 enrollees. Application Market and Deployment (Concluded) Iris-scan is set to grow substantially through 2007 and beyond. Iris-scan offers low false match rates and hands-free operation, and is the only viable alternative to fingerprint technologies in 1:N applications where a single record must be located. Iris-scan revenues are projected to grow from $16.2m in 2002 to $210.2m in 2007. Iris-scan revenues are expected to comprise approximately 5% of the entire biometric market. Strengths of Iris Scan Resistance to False Matching - 1 in 1,200,0000 approx. Stability of Characteristic over lifetime - Characteristics formed pre-birth; changed only by injury Can be used for both logical and physical access Weaknesses of Iris Scan Difficult to use - Acquisition systems are not straightforward - User must be positioned correctly False nonmatching and failure to enroll - False nonmatch rates need to improve - Works with smaller databases - Difficult to capture the images User discomfort - Users reluctant to capture their iris Acquisition devices are mostly proprietary Research Directions Improve False Nonmatch rates Better performance for larger databases Capture images when user is wearing glasses Techniques for very dark eyes Technology Comparison Method Coded Pattern Misidentification rate Security Iris Recognition Iris pattern 1/1,200,000 Fingerprinting Fingerprints 1/1,000 Hand Shape Size, length and thickness of hands Facial Recognition Outline, shape and distribution of eyes and nose 1/700 1/100 Signature Shape of letters, writing order, pen pressure 1/100 Voiceprinting Voice characteristics 1/30 Summary Low failure rate: 1 in 1,200,000 approx Usable for highly secure applications Need better acquisition techniques Need better performance Introduction to Biometrics Dr. Bhavani Thuraisingham The University of Texas at Dallas Information Relevant to the Project – Version 2 September 28, 2005 Outline of the Unit Project Information Some Details of the Project Project Information In this project, you are asked to do the two following tasks: Recognize one’s Face. Recognize one’s face with different poses (i.e. straight, left, right and up). Project Information (Continued) PART1 For this experiment you will use neural network package given in the code subdirectory in location http://www.cs.cmu.edu/afs/cs.cmu.edu/user/avrim/www/M L94/face_homework.html - For training and testing, you will use the face images that are listed in the trainset subdirectory in - http://www.cs.cmu.edu/afs/cs.cmu.edu/user/avrim/www/M L94/face_homework.html - The face images that are listed in the trainset subdirectory are given in the faces subdirectory in the same above location. Project Information (Continued) PART2 (OPTIONAL) You will also use k-nearest neighborhood using the same dataset and compare its performance with neural network. Project Information (Continued) RECOMMENDED You don’t need to do significant amounts of coding for part1 of the project. Only you need to make small changes in the files, imagenet.c and facetrain.c given in the code subdirectory in the location - http://www.cs.cmu.edu/afs/cs.cmu.edu/user/avrim/www/M L94/face_homework.html For training your datasets in part 1, it will take time. It is recommended that you read the materials in the above location thoroughly ( particularly in assignment document, section 6,”Documantations” ) and start it earlier. Project Information (Concluded) For part2 (OPTIONAL), you need some extensive coding. You will use k-nearest neighborhood instead of back propagation in the given code. Some Details of the Project FACE IMAGES The image data can be found in the faces subdirectory in - http://www.cs.cmu.edu/afs/cs.cmu.edu/user/avrim/www/M L94/face_homework.html This subdirectory contains 20 subdirectories, one for each person who volunteered for the photo shoot, named by userid. Each of these subdirectories contains several versions of the face images. For images the following naming convention is used: <userid>_<pose>_< expression>_<eyes>_<scale>.pgm Some Details of the Project (Continued) For further details see the assignment document, section 2 in the following link. - http://www.cs.cmu.edu/afs/cs.cmu.edu/user/avrim/www/M L94/face_homework.html Some Details of the Project (Continued) HOW TO VIEW THE FACE IMAGES You will need View.java and View.class to view the images. These two files will be placed on my web site by tomorrow View.java handles a variety of image formats, including the PGM format in which the face images are stored. To start View, just type on the command line: java View ImageInput Here ImageInput corresponds to the image you want to view Some Details of the Project (Continued) THE NEURAL NETWORK AND IMAGE ACCESS CODE You can have C code for a three layer fully connected feed forward neural network which uses the back propagation algorithm to tune its weights. You can also have the top level program ( facetrain.c ) for training and recognition, as a skeleton for you to modify. The code is located in the code directory in the following location - http://www.cs.cmu.edu/afs/cs.cmu.edu/user/avrim/www/M L94/face_homework.html Copy all the files in this area to your UTD home directory, and type make. Some Details of the Project (Continued) When the compilation is done, you should have one executable program : facetrain. facetrain takes lists of image file as input and uses these as training and test sets for a neural network. facetrain can be used for training and/or recognition, and it also has the capability to save networks to file. facetrain outputs a number of performance measures in the output file at the end of each epoch, in the folllowing format <epoch> <delta> <trainperf> <trainerr> <t1perf> <t1err> <t2perf> <t2err> Some Details of the Project (Continued) For further details see in the subdirectory assignment documents, section 4 & 6.2 in the following link - http://www.cs.cmu.edu/afs/cs.cmu.edu/user/avrim/www/M L9/face_homework.html Some Details of the Project (Continued) ASSIGNMENT Part1 1. Copy straight_train.list, straight_test1.list, straight_test2.list, all_train.list, all_test1.list, all_test2.list in your home directory to obtain the training and test set data for this assignment from the following trainset link in http://www.cs.cmu.edu/afs/cs.cmu.edu/user/avrim/www/ ML94/face_homework.html - 2. Train it with the default learning parameter settings (learning rate 0.3, momentum 0.3) for 75 epochs, with the following command. facetrain –n you.net –t straight_train.list -1 straight_test1.list -2 straight_test2.list –e 75 Some Details of the Project (Continued) you.net is the name of the network file which will be saved when training is finished. straight_train. list, straight_test1.list, and straight_test2.list are text files which specify the training set (70 examples) and two test sets (32 and 50 examples), respectively. This command creates and trains your net on randomly chosen sample of 70 of the 152 “straight” images, and test it on the remaining 32 and 50 randomly chosen images, respectively. Report your train/test performance and error as a function of epochs. If you had stopped training when the performance on test1 leveled off, what would the performance have been on test2? And vice versa? Some Details of the Project (Continued) 3. Implement a face recognizer; i.e. implement a neural net which, when given an image as input, indicates who is in the image. To do this, you will need to implement a different output encoding (since you must now be able to distinguish among 20 peoples). Describe your output encoding. (Hint: leave learning rate and momentum at 0.3, and use 20 hidden units). Train the network for 100 epochs: facetrain –n face.net –t straight_train.list -1 straight_test1.list -2 straight_test2.list –e 100 Report your train/test performance and error as a function of epochs. If you had stopped training when the performance on test1 leveled off, what would the performance have been on test2? And vice versa? Some Details of the Project (Continued) 4. Implement a pose recognizer; i.e. implement a neural net which, when given an image as input, indicates whether the person in the image is looking straight ahead, up, to the left, or to the right. You will also need to implement a different output encoding for this task. Describe your output encoding. (Hint: leave learning rate and momentum at 0.3, and use 6 hidden units). Train the network for 100 epochs: facetrain –n pose.net –t all_train.list -1 all_test1.list -2 all_test2.list –e 100 Report your train/test performance and error as a function of epochs. If you had stopped training when the performance on test1 leveled off, what would the performance have been on test2? And vice versa? Some Details of the Project (Continued) What changes you should make in the code. You will need to modify the routine load_target in code imagenet.c to setup appropriate target vectors for the output encodings you choose, when implementing the face recognizer and the pose recognizer. You will need to modify the code facetrain.c to change network sizes and learning parameters, both of which are trivial changes. You will need to modify the two performance evaluation routines performance_on_imagelist() and evaluate_performance() in code facetrain.c ,when implementing the face recognizer and the pose recognizer. 5. Some Details of the Project (Concluded) 6. For further assistance see in assignment document, section 5-3, 5-6, 5-8 and 6 in the following link http://www.cs.cmu.edu/afs/cs.cmu.edu/user/avrim/www/ML9/f ace_homework.html Part2 (OPTIONAL) Use k-nearest neighborhood to do all the tasks above. Here you will determine the best k value.
© Copyright 2024 Paperzz