The goal of this course is to become familiar with research advances and techniques in the fast developing field of biometric person authentication. In the first few lectures we will review important topics in pattern recognition. We will then study general topics on biometrics: biometric traits, system errors and performance, biometric databases and indexing, and security.We will explore in detail the commonly used biometric modalities of fingerprints and faces.
The workload will include three projects - two mid-term projects and one major term project. In the term project you will create a fingerprint or face biometric matcher. Grading will be as follows: project 1 (20%), project 2 (30%), term project (40%) and class participation (10%). Projects 1 and 2 require you to turn in a report, and the term project requires you to make a 30 min. presentation in addition to turning in a project report. A few short quizzes will be given on random days; their grades will be included in class participation.
This course expands on the material presented in following courses: CSE 555 (Pattern Recognition), CSE 573 (Computer Vision and Image Processing), CSE 574 (Machine Learning). We expect the students to have taken at least two of these courses prior to taking Biometrics course. The term project involves building new or modification of existing biometric systems (written in C++), so good programming skills are a must. For example, you have to be able to incorporate some machine learning method (such as SVM or neural network) from existing library in your code, use OpenCV library, design a machine learning training/testing procedure or build a user interface.
There is no required text for this course. The recommended reference books are:
R. Duda, P. Hart and D. Stork, "Pattern Classification"
R. Bolle et al., "Guide to Biometrics"
D. Maltoni et al., "Handbook of Fingerprint Recognition"
N.K. Ratha and V.Govindaraju (Eds.), "Advances in Biometrics: Sensors, Systems
A. Jain et al. (Eds.), "Biometrics. Personal Identification in a Networked
Kanade Expression Dataset: This dataset consists of approximately
2000 images from over 200 subjects. Subjects were instructed by an experimenter
to perform a series of 23 facial displays that included single action
units and combinations of action units. Images are digitized into 640
by 480 or 490 pixel arrays with 8-bit precision for grayscale values.
Face Dataset: This dataset consists of over 4,000 color images corresponding
to 126 people's faces (70 men and 56 women). Images feature frontal view
faces with different facial expressions, illumination conditions, and
occlusions (sun glasses and scarf). Each person participated in two sessions,
separated by two weeks (14 days) time. The same pictures were taken in
SRE 2000: This dataset consists of 10,328 single channel SPHERE files
encoded in 8-bit mµ-law containing a total of approximately 4.31 Gbytes
of data covering 148.9 hours of audio.
SRE 2001: This dataset consists of 2,350 compressed speech files,
all of which are in SPHERE format. The files are compressed and encoded
in one channel 8-bit mµ-law, for a total of 548.7 Mbytes, or 26 hours
of audio data.
SRE 2002: This database consists of 9,153 speech files (6,098 at 8
KHz and 3,055 at 16KHz), all of which are in SPHERE format, for a total
of ~156 hours.
NIST SD-14: This
dataset consists of 27,000 pairs of segmented 8-bit grayscale fingerprint
images each of size 832x768 pixels scanned at 19.7 pixels per mm. The
database has been classified using the National Crime Information Center
(NCIC) fingerprint classes given by the FBI.
This dataset consists of four separate sets of print, two sets (DB1 and
DB2) collected using optical technology sensors, one set (DB3) collected
using a capacitive sensor and one set (DB4) consisting of synthetically
generated prints. Each set contains 110 fingers with 8 prints per finger
for a total of 880 prints. DB1 consists of images of size 388x374 pixels
scanned at 500 dpi. DB2 consists of images of size 296x560 pixels scanned
at 569 dpi. DB3 consists of images of size 300x300 pixels scanned at 500
This dataset is similar to the one provided for FVC 2002 except that DB3
collected using a sweeping thermal scanner and all prints were perturbed
deliberately to increase the matching complexity.
CASIA Gait Dataset:
This dataset is composed of three subsets. Dataset A includes 20 subjects.
Each subject has 12 image sequences, 4 sequences for each of the three
directions, i.e. parallel, 45 degrees and 90 degrees to the image plane.
The length of each sequence is not identical for the variation of the
walker's speed. Dataset B is a large multiview gait database. There are
124 subjects, and the gait data was captured from 11 views. Three variations,
namely view angle, clothing and carrying condition changes, are separately
considered. Besides the video files, human silhouettes extracted from
the video files are also available. Dataset C was collected by an infrared
(thermal) camera. It contains 153 subjects and takes into account four
walking conditions: normal walking, slow walking, fast walking, and normal
walking with a bag. The videos were all captured at night. The entire
dataset size is ~10GB for the video files and ~700MB for just the image
This dataset includes three subsets - CASIA-IrisV3-Interval, CASIA-IrisV3-Lamp,
CASIA-IrisV3-Twins. The dataset contains a total of 22,051 iris images
from more than 700 subjects. All iris images are 8 bit gray-level JPEG
files, collected under near infrared illumination.
SVC 2004: This
dataset consists of two separate signature databases. The signature data
for the first task contain coordinate information only, but the signature
data for the second task also contain additional information including
pen orientation and pressure. Each database has 100 sets of signature
data. Each set contains 20 genuine signatures from one signature contributor
and 20 skilled forgeries from at least four other contributors.