Jason Thornton

A Bayesian Approach to Deformed Pattern Matching of Iris Images

We describe a general probabilistic framework for matching patterns that experience in-plane nonlinear deformations, such as iris patterns. Given a pair of images, we derive a maximum a posteriori probability (MAP) estimate of the parameters of the relative deformation between them. Our estimation process accomplishes two things simultaneously: it normalizes for pattern warping and it returns a distortion-tolerant similarity metric which can be used for matching two nonlinearly deformed image patterns. The prior probability of the deformation parameters is specific to the pattern-type and, therefore, should result in more accurate matching than an arbitrary general distribution. We show that the proposed method is very well suited for handling iris biometrics, applying it to two databases of iris images which contain real instances of warped patterns. We demonstrate a significant improvement in matching accuracy using the proposed deformed Bayesian matching methodology. We also show that the additional computation required to estimate the deformation is relatively inexpensive, making it suitable for real-time applications

Biometric verification with correlation filters

Using biometrics for subject verification can significantly improve security over that of approaches based on passwords and personal identification numbers, both of which people tend to lose or forget. In biometric verification the system tries to match an input biometric (such as a fingerprint, face image, or iris image) to a stored biometric template. Thus correlation filter techniques are attractive candidates for the matching precision needed in biometric verification. In particular, advanced correlation filters, such as synthetic discriminant function filters, can offer very good matching performance in the presence of variability in these biometric images (e.g., facial expressions, illumination changes, etc.). We investigate the performance of advanced correlation filters for face, fingerprint, and iris biometric verification.

Iris verification using correlation filters

Iris patterns are believed to be an important class of biometrics suitable for subject verification and identification applications. Earlier methods proposed for iris recognition were based on generating iris codes from features generated by applying Gabor wavelet processing to iris images. Another approach to image recognition is the use of correlation filters. Correlation filter methods differ from many image-based recognition approaches in that two-dimensional Fourier transforms of the images are used in this approach. In correlation filter methods, normal variations in an authentic iris image can be accommodated by designing a frequency-domain array (called a correlation filter) that captures the consistent part of iris images while deemphasizing the varying parts. Correlation filters also offer other benefits such as shift-invariance, graceful degradation and closed-form solutions. In this paper, we discuss the basics of correlation filters and show how they can be used for iris verification.

Person attribute search for large-area video surveillance

This paper describes novel video analytics technology which allows an operator to search through large volumes of surveillance video data to find persons that match a particular attribute profile. Since the proposed technique is geared for surveillance of large areas, this profile consists of attributes that are observable at a distance (including clothing information, hair color, gender, etc) rather than identifying information at the face level. The purpose of this tool is to allow security staff or investigators to quickly locate a person-of-interest in real time (e.g., based on witness descriptions) or to speed up the process of video-based forensic investigations. The proposed algorithm consists of two main components: a technique for detecting individual moving persons in large and potentially crowded scenes, and an algorithm for scoring how well each detection matches a given attribute profile based on a generative probabilistic model. The system described in this paper has been implemented as a proof-of-concept interactive software tool and has been applied to different test video datasets, including collections in an airport terminal and collections in an outdoor environment for law enforcement monitoring. This paper discusses performance statistics measured on these datasets, as well as key algorithmic challenges and useful extensions of this work based on end-user feedback.1

Robust iris recognition using advanced correlation techniques

The iris is considered one of the most reliable and stable biometrics as it is believed to not change significantly during a persons lifetime. Standard techniques for iris recognition, popularized by Daugman, apply Gabor wavelet analysis for feature extraction. In this paper, we consider an alternative method for iris recognition, the use of advanced distortion-tolerant correlation filters for robust pattern matching. These filters offer two primary advantages: shift invariance, and the ability to tolerate within-class image variations. The iris images we use in our experiments are from the CASIA database and also from an iris database we collected at CMU. In this paper, we perform automatic segmentation of the iris (which surrounds the pupil) from the rest of the eye, normalizing for scale and pupil dilation. We then use these segmented iris images to compare the recognition performance of various methods, including Gabor wavelet feature extraction, to correlation filters.

An Evaluation of Iris Pattern Representations

The success of an iris recognition algorithm is partially dependent upon the iris pattern representation computed during feature extraction. Some algorithms in the literature represent iris texture by applying bandpass Alter banks to the segmented iris region. However, the selection of a bandpass filter form, as well as the particular instantiations of that form, are often presented as arbitrary choices. In this paper, we evaluate multiple filter candidates and compare the discriminative information provided by their responses. Discrimination is measured on a set of reference iris images drawn from multiple datasets. After preliminary analysis of the best filter type, we conduct iterative optimization over the parameters of a filter bank to search for the best possible iris pattern representation. Finally, we give some sample recognition results using the selected filter bank.

Wavelet packet correlation methods in biometrics

We introduce wavelet packet correlation filter classifiers. Correlation filters are traditionally designed in the image domain by minimization of some criterion function of the image training set. Instead, we perform classification in wavelet spaces that have training set representations that provide better solutions to the optimization problem in the filter design. We propose a pruning algorithm to find these wavelet spaces by using a correlation energy cost function, and we describe a match score fusion algorithm for applying the filters trained across the packet tree. The proposed classification algorithm is suitable for any object-recognition task. We present results by implementing a biometric recognition system that uses the NIST 24 fingerprint database, and show that applying correlation filters in the wavelet domain results in considerable improvement of the standard correlation filter algorithm.

Graphical Model Approach to Iris Matching Under Deformation and Occlusion

Template matching of iris images for biometric recognition typically suffers from both local deformations between the template and query images and large occlusions from the eyelid. In this work, we model deformation and occlusion as a set of hidden variables for each iris comparison. We use afield of directional vectors to represent deformation and a field of binary variables to represent occlusion. We impose a probability distribution on these fields using a lattice-type undirected graphical model, in which the graph edges represent interdependencies between neighboring iris regions. Gabor wavelet-based similarity scores and intensity statistics are used as observations in the model. Loopy belief propagation is applied to estimate the conditional distributions on the hidden variables, which are in turn used to compute final match scores. We present underlying theory as well as experimental results from both the CASIA iris database and the database provided for the iris challenge evaluation (ICE). We show that our proposed method significantly improves recognition accuracy on these datasets over existing methods.

Using composite correlation filters for biometric verification

Biometric verification refers to the process of matching an input biometric to stored biometric information. In particular, biometric verification refers to matching the live biometric input from an individual to the stored biometric template of that individual. Examples of biometrics include face images, fingerprint images, iris images, retinal scans, etc. Thus, image processing techniques prove useful in biometric recognition. In particular, composite correlation filters have proven to be effective. In this paper, we will discuss the application of composite correlation filters to biometric verification.

Linear shift-invariant maximum margin SVM correlation filter

Advanced correlation filters are effective for recognizing distorted images of a particular class. Most correlation filter designs are based on optimization criteria that lead to a closed form filter solution. We remove this restriction of a closed form solution and introduce a new filter design approach, based on a margin of separation maximization formulated as a linear support vector machine (SVM). The resulting SVM classifier is of the form of a correlation filter which has some attractive attributes, such as linearity and shift-invariance (properties that traditional SVM classifiers lack). We also show that our proposed SVM correlation filter offers built-in noise tolerance, which is valuable for any recognition task where noise can be present. More importantly, we demonstrate that we can achieve good generalization using only a single image for training. We compare our proposed filter design to popular advanced correlation filter designs and show the increase in performance of our proposed method by testing on two well known face databases (CMU-AMP lab facial expression database and the CMU-PIE illumination dataset consisting of faces of 65 people).