Speaker: Rohan Nadgir
Date: 4/21/2006, Friday (Andrews Conf Room)
Time: 1:00-2:00pm


Abstract: Biometric sensor interoperability refers to the ability of a system to compensate for the variability introduced in the biometric data of an individual due to the deployment of different sensors. We demonstrate that a simple non-linear calibration scheme, based on Thin Plate Splines (TPS), is sufficient to facilitate sensor interoperability in the context of fingerprints. In the proposed technique, the variation between the images acquired using two different sensors is modeled using non-linear distortions. Experiments indicate that the proposed calibration scheme can significantly improve inter-sensor matching performance.

Speaker: Rajiv Mukherjee
Date: 3/24/2006, Friday (801 ESB)
Time: 12:45-1:45p


Abstract: Fingerprints based biometric systems are used in two ways: verification and identification. The identification problem is more complex given the large database size. To combat the problem of database size and variability in fingerprint from the same user, classification and indexing techniques are used. Indexing is a technique used to significantly reduce the number of prospective hypotheses to be considered by the matching algorithm. The speaker will describe traditional indexing techniques and will discuss two new indexing techniques. The first technique is based on alloting an index to the fingerprint itself. The other is based on a two-pronged elimination scheme.

Speaker: Simona Crihalmeanu
Date: 3/10/2006, Friday (Brown Conf Room)
Time: 1:00-2:00pm


Abstract: Among different biometric modalities, the iris biometric is considered to be reliable due to its variability across individuals and the perceived stability over time. With the use of high resolution cameras, iris recognition technology has improved considerably especially in the context of frontal images of the eye. However, in non-ideal situations (such as off-angle iris acquisition), the performance of traditional iris algorithms is expected to degrade. When an image of the eye is captured from multiple angles, there are details of the sclera that are exposed. Our approach is to establish if the structure of the veins from the sclera complies with the seven desirable properties of any biometric characteristic: universality, uniqueness, permanence, collectability, performance, acceptability, and spoof-proof. In this talk we discuss the salient features that support the notion of using sclera veins as a biometric.

Speaker: Samir Shah
Date: 2/23/2006, Friday (Brown Conf Room)
Time: 2:45-3:45pm


Abstract:

Speaker: Phani Ivatury
Date: 2/17/2005, Thursday (Brown Conf Room)
Time: 4:00 - 5:00 pm


Abstract: In this talk, I would like to discuss some of the techniques I have encountered, worked on, in due course of my research on Facial Feature Localisation. In particular, I would be summarising the papers Deformable templates (by Yuille et al.) and Snakes (by Kass et al.).

Speaker: Chris Boyce
Date: 2/10/2005, Thursday (Brown Conf Room)
Time: 4:00 - 5:00 pm


Abstract: To understand the complications and complexity of imaging the iris, an understanding of iris anatomy is first pivotal. My talk entitled Infrared Iris Imagery explains the complex anatomy of the anterior portion of the eye, the iris' physiological properties, and some abnormalities that can occur in the iris and affect its recognition performance. I will then discuss some of the physics of electromagnetic radiation related to reflection and absorption in the iris, and describe some of the systems (cameras, lenses, filters) that can be used for iris imagery. I will also discuss some of the current imaging problems associated with the iris. I will conclude with a discussion on the absorption of various wavelengths of near infrared light and their effects on various pigmented irises (blue, green, brown).

Speaker: Rohan Nadgir
Date: 2/3/2005, Thursday (Brown Conf Room)
Time: 4:00 - 5:00 pm


Abstract: The performance of a fingerprint system is closely associated with the quality of the participating fingerprints. The process of minutiae detection is impacted by the quality of the ridge information present in the image. A poor quality image would eventually lead to increased FRR during recognition and increased FTE during enrollment. Thus, we need to assess fingerprint quality before extracting and storing the template. In my talk, I will present an overview of techniques that have been presented in the literature toward quantifying fingerprint image quality.

Speaker: Jidnya Shah
Date: 1/27/2005, Thursday (Brown Conf Room)
Time: 4:00 - 5:00 pm


Abstract: Most minutiae-based fingerprint matching systems store minutiae information (known as a template) from the fingerprint image of a user. Since this fingerprint template does not store information about singularities, ridge details, and class of the fingerprint it has been always considered safe to store them in database. We show that minutiae information can reveal substantial details such as the orientation field and the class of the associated fingerprint that can potentially be used to reconstruct the original fingerprint image. The proposed technique utilizes minutiae triplet information to estimate the orientation map of the parent fingerprint. The estimated orientation map is observed to be remarkably consistent with the underlying ridge flow. A fingerprint pattern classifier that utilizes minutiae information alone to infer the class of the fingerprint is designed. Each fingerprint is represented by a feature vector based on its minutiae properties. The classification error rate on NIST 4 database is 18%. This indicates that the seemingly random minutiae distribution of a fingerprint can reveal important class information. Furthermore, contrary to what has been claimed by several minutiae-based fingerprint system vendors, a dynamic algorithm to reconstruct fingerprint image from a minutiae template is proposed.

Speaker: Sarvesh Makthal
Date: 1/20/2005, Thursday (Brown Conf Room)
Time: 4:00 - 5:00 pm


Abstract: A number of iris recognition algorithms have been proposed and developed in the past few years with some of them being implemented commercially too. While all of these methods claim very accurate identification, most of them have never been tested on very large databases. The largest known database has about 350,000 iris images, but is private. The talk will introduce a synthetic iris generation method based on Markov random field modeling. The iris is perceived as a texture image and a fast texture synthesis method is used to generate new synthetic iris images. Image data from real world iris images is used as input to the algorithm for synthetic image generation. It will also be shown that this procedure can be significantly accelerated using tree-structured vector quantization. Further, clustering experiments show that the iris texture is significantly different from the regular stochastic textures and that the synthetic iris generated here can be classified as iris texture.

Speaker: Dr. Vijayakumar Bhagavatula, Carnegie Mellon University
Date: 9/29/2004, Wednesday (280 MRB)
Time: 2:30 - 3:30 pm


Abstract: In the increasingly e-commerce oriented society, verifying a user's identity is critical for carrying out financial and other transactions with trust. Most current authentication systems are password based making them susceptible to problems such as forgetting the password and passwords being stolen. One way to overcome these problems is to employ biometrics (e.g., fingerprints, face images, iris images, palm prints, etc.) for authentication. This talk will provide an overview of the research in methods to authenticate a person's identity based on their biometrics, being carried out at Carnegie Mellon's CyLab. In particular, the application of correlation filters to verify the identity based on face images, fingerprint images and iris images will be discussed. Correlation filters are designed in the spatial frequency domain and offer several advantages such as shift-invariance, closed-form designs and graceful degradation. Although the focus of this talk is on verification, we will also show results of applying these methods to the task of face identification. A laptop-based real-time demo of face verification will also be presented.

Speaker: Simona Crihalmeanu
Date: 8/13/2004, Friday (Brown Conf Room)


Abstract: This talk will discuss liveness detection measures for fingerprint systems based on the work done by Stephanie Schuckers' research group.

Speaker: Diogo Pereira
Date: 7/23/2004, Friday (Brown Conf Room)


Abstract: This study investigates a (IR) face recognition system using an uncooled IR camera. A computer-based image collection set-up was designed and used to create a small database of 420 facial images, from 14 volunteers. Manual and automated facial image cropping routines were implemented. Two linear approaches (PCA and LDA) for dataset dimension reduction and classification were implemented and the resulting classification performances compared. Results show that the best PCA-based average classification performance is 92.22% while the LDA-based classification performance is 99.40%.

Speaker: Sruti Ramnath
Date: 7/08/2004, Thursday (ESB 207)


Abstract: This presentation discusses saccadic eye movements and their importance, specifically for scene representation in human memory. In this regard the following paper will be discussed: Henderson, J. M., & Hollingworth, A. (2003). Eye movements and visual memory: Detecting changes to saccade targets in scenes. Perception & Psychophysics, 65, 58-71. The objective of the paper is to understand the degree of detail of scene representations that is retained in memory. The presentation concludes with a look at what can be further explored in the study of saccades using image processing and computer vision techniques.

Speaker: Jidnya Shah
Date: 7/02/2004, Friday (Brown Conf Room)


Abstract: All existing fingerprint classification schemes use fingerprint image along with its features like ridge line flow, singularities, orientation image or Gabor filter responses for classification. We propose a novel idea of classifying fingerprints using minutiae information. We support our hypothesis by referring to the use of minutiae distribution in fingerprint individuality models proposed by various forensic experts in the literature. We have designed a classifier which assigns a class to a fingerprint based on its minutiae features. The preliminary results show that there is an indispensable relation between minutiae locations, their orientations and the class of the fingerprint. We claim that the knowledge about minutiae location and their orientation alone is sufficient for classifying fingerprints. We also discuss the possibility of reconstructing fingerprint images from the seemingly random distribution of its minutiae points using Gabor-like filters.

Speaker: Rohin Govindarajan
Date: 6/18/2004, Friday (Brown Conf Room)


Abstract: Multimodal biometric systems utilize the evidence presented by multiple biometric sources (e.g., face and Fingerprint, multiple Fingers of a user, multiple impressions of a single Finger, etc.) in order to determine or verify the identity of an individual. Information from multiple sources can be consolidated in three distinct levels: (i) feature extraction level; (ii) match score level; and (iii) decision level. While fusion at the match score and decision levels have been extensively studied in the literature, fusion at the feature level is a relatively understudied problem. A novel technique to perform fusion at the feature level by considering two biometric modalities - face and hand geometry, is presented in this talk. This technique utilizes the fused feature vectors of face and hand geometry in order to improve the performance of a multimodal biometric system. Also, a new distance measure, called the Thresholded Absolute Distance is discussed, which helps increase the system's robustness towards noise. Finally, different techniques to consolidate information at the match score level with that at the feature level, are proposed. These techniques help in further enhancing the performance of the multimodal biometric system and help in finding an approximate upper bound on its performance. An unique application of multibiometrics in applications requiring group consensus titled `Biometrics with group consensus' is proposed. The proposed techniques help in consolidating biometric information pertaining to multiple users in order to authenticate them as a group.

Speaker: Sarvesh Makthal
Date: 6/11/2004, Friday (Brown Conf Room)


Abstract: An introduction to texture analysis and its importance in machine vision will be given. Various methods used for texture analysis and feature extraction will be listed under different categories : Statistical, Geometrical, Model-based and Signal Processing-based methods. Popular approaches in each category will be discussed in brief to give the audience an idea of the properties exploited in each category. The need for texture analysis will then be highlighted with specific examples of where texture analysis is used in computer vision (Image segmentation, Iris recognition, etc.).