The Experts below are selected from a list of 1374 Experts worldwide ranked by ideXlab platform

Carolina Parada - One of the best experts on this subject based on the ideXlab platform.

  • convolutional neural networks for small footprint keyword spotting
    Conference of the International Speech Communication Association, 2015
    Co-Authors: Tara N Sainath, Carolina Parada
    Abstract:

    We explore using Convolutional Neural Networks (CNNs) for a small-footprint keyword spotting (KWS) task. CNNs are attractive for KWS since they have been shown to outperform DNNs with far fewer parameters. We consider two different applications in our work, one where we limit the number of multiplications of the KWS system, and another where we limit the number of parameters. We present new CNN architectures to address the constraints of each applications. We find that the CNN architectures offer between a 27-44% relative improvement in False Reject Rate compared to a DNN, while fitting into the constraints of each application.

  • INTERSPEECH - Convolutional Neural Networks for Small-Footprint Keyword Spotting
    2015
    Co-Authors: Tara N Sainath, Carolina Parada
    Abstract:

    We explore using Convolutional Neural Networks (CNNs) for a small-footprint keyword spotting (KWS) task. CNNs are attractive for KWS since they have been shown to outperform DNNs with far fewer parameters. We consider two different applications in our work, one where we limit the number of multiplications of the KWS system, and another where we limit the number of parameters. We present new CNN architectures to address the constraints of each applications. We find that the CNN architectures offer between a 27-44% relative improvement in False Reject Rate compared to a DNN, while fitting into the constraints of each application.

Tara N Sainath - One of the best experts on this subject based on the ideXlab platform.

  • convolutional neural networks for small footprint keyword spotting
    Conference of the International Speech Communication Association, 2015
    Co-Authors: Tara N Sainath, Carolina Parada
    Abstract:

    We explore using Convolutional Neural Networks (CNNs) for a small-footprint keyword spotting (KWS) task. CNNs are attractive for KWS since they have been shown to outperform DNNs with far fewer parameters. We consider two different applications in our work, one where we limit the number of multiplications of the KWS system, and another where we limit the number of parameters. We present new CNN architectures to address the constraints of each applications. We find that the CNN architectures offer between a 27-44% relative improvement in False Reject Rate compared to a DNN, while fitting into the constraints of each application.

  • INTERSPEECH - Convolutional Neural Networks for Small-Footprint Keyword Spotting
    2015
    Co-Authors: Tara N Sainath, Carolina Parada
    Abstract:

    We explore using Convolutional Neural Networks (CNNs) for a small-footprint keyword spotting (KWS) task. CNNs are attractive for KWS since they have been shown to outperform DNNs with far fewer parameters. We consider two different applications in our work, one where we limit the number of multiplications of the KWS system, and another where we limit the number of parameters. We present new CNN architectures to address the constraints of each applications. We find that the CNN architectures offer between a 27-44% relative improvement in False Reject Rate compared to a DNN, while fitting into the constraints of each application.

Stephanie Schuckers - One of the best experts on this subject based on the ideXlab platform.

  • WIFS - Effects of text filtering on authentication performance of keystroke biometrics
    2016 IEEE International Workshop on Information Forensics and Security (WIFS), 2016
    Co-Authors: Jiaju Huang, Stephanie Schuckers, Shambhu Upadhyaya
    Abstract:

    Free text keystroke dynamics is a behavioral biometric that has the strong potential to offer unobtrusive and continuous user authentication. In free-text keystroke biometrics, users are free to type whatever they want to while still being authenticated. However, not all keystrokes from a user exhibit the same quality of stable patterns that can be used to differentiate them from others. The “unstable” keystrokes may originate from such activities as when the user is playing a computer game, or other sources of noisy or “gibberish” text. Our hypothesis is that some forms of text negatively impact keystroke dynamics-based user authentication, and thus, should be filtered out. This study investigates the impact of gibberish text on authentication performance through locating and removing gibberish text, and comparing the difference in authentication performance before and after the removal. We confirm the positive effect of text filtering on authentication performance, especially on the reduction of the False Reject Rate.

  • integrating a wavelet based perspiration liveness check with fingerprint recognition
    Pattern Recognition, 2009
    Co-Authors: Aditya Abhyankar, Stephanie Schuckers
    Abstract:

    It has been shown that fingerprint scanners can be deceived very easily, using simple, inexpensive techniques. In this work, a countermeasure against such attacks is enhanced, that utilizes a wavelet based approach to detect liveness, integRated with the fingerprint matcher. Liveness is determined from perspiration changes along the fingerprint ridges, observed only in live people. The proposed algorithm was applied to a data set of approximately 58 live, 50 spoof and 28 cadaver fingerprint images captured at 0 and 2s, from each of three different types of scanners, for normal conditions. The results demonstRate perfect separation of live and not live for the normal conditions. Without liveness module the commercially available verifinger matcher is shown to give equal error Rate (EER) of 13.85% where False Reject Rate is calculated for genuine-live users and False accept Rate is for genuine-not live, imposter-live and imposter-not live. The integRated system of fingerprint matcher and liveness module reduces EER to 0.03%. Results are also presented for moist and dry fingers simulated by glycerin and acetone, respectively. The system is further tested using gummy fingers and various delibeRately simulated conditions including pressure change and adding moisture to the spoof to analyze the strength of the liveness algorithm.

Arun Ross - One of the best experts on this subject based on the ideXlab platform.

  • A Multiscale Sequential Fusion Approach for Handling Pupil Dilation in Iris Recognition
    Human Recognition in Unconstrained Environments, 2017
    Co-Authors: Raghunandan Pasula, Simona Crihalmeanu, Arun Ross
    Abstract:

    Abstract Pupil dilation is shown to degrade iris recognition accuracy (Hollingsworth et al., 2009, [7] ) by increasing the Hamming distance between genuine samples thereby increasing the False Reject Rate (FRR). Dilation of pupil due to illumination changes or drugs results in complex deformation of the iris texture. In this work, a database to study the effect of pupil dilation is acquired. The adverse effect of pupil dilation on iris recognition is demonstRated using this database. A novel fusion scheme that sequentially fuses IrisCode bit information at different scales is proposed to reduce the effect of non-linear deformation. The experimental results show that the proposed method improves the matching accuracy on both the pupil dilation data as well as other non-ideal iris data.

  • Analysis of User-specific Score Characteristics for Spoof Biometric Attacks
    2016
    Co-Authors: Ajita Rattani, Norman Poh, Arun Ross
    Abstract:

    Several studies in biometrics have confirmed the exis-tence of user-specific score characteristics for genuine and zero-effort impostor score distributions. As an important consequence, biometric users contribute disproportionately to the FRR (False Reject Rate) and FAR (False accept Rate) of the system. This phenomena is also know as the Dod-dington zoo effect. Recent studies indicate the vulnerability of unimodal and multibiometric systems to spoof attacks. The aim of this study is to analyze the score characteris-tics for spoof attacks. Such an analysis will 1) help im-prove our understanding of the Doddington zoo effect under spoof attacks; and 2) allow us to design biometric classi-fiers that are more robust to such attacks. The contributions of this paper are as follows: a) examining the existence of user-specific score characteristics for spoof attacks and b) analyzing the correlation between user-specific score char-acteristics obtained on genuine (as well as zero-effort im-postor) and non zero-effort impostor (spoof) score distribu-tions. Experiments conducted on the LivDet09 spoofed fin-gerprint database confirms that biometric user-groups ex-hibit different degrees of vulnerability to spoof attacks as well. Further, modeRate negative correlation may exist be-tween users who are difficult to recognize and their vulner-ability to spoof attacks. 1

  • CVPR Workshops - Analysis of user-specific score characteristics for spoof biometric attacks
    2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, 2012
    Co-Authors: Ajita Rattani, Arun Ross
    Abstract:

    Several studies in biometrics have confirmed the existence of user-specific score characteristics for genuine and zero-effort impostor score distributions. As an important consequence, biometric users contribute disproportionately to the FRR (False Reject Rate) and FAR (False accept Rate) of the system. This phenomena is also know as the Doddington zoo effect. Recent studies indicate the vulnerability of unimodal and multibiometric systems to spoof attacks. The aim of this study is to analyze the score characteristics for spoof attacks. Such an analysis will 1) help improve our understanding of the Doddington zoo effect under spoof attacks; and 2) allow us to design biometric classifiers that are more robust to such attacks. The contributions of this paper are as follows: a) examining the existence of user-specific score characteristics for spoof attacks and b) analyzing the correlation between user-specific score characteristics obtained on genuine (as well as zero-effort impostor) and non zero-effort impostor (spoof) score distributions. Experiments conducted on the LivDet09 spoofed fingerprint database confirms that biometric user-groups exhibit different degrees of vulnerability to spoof attacks as well. Further, modeRate negative correlation may exist between users who are difficult to recognize and their vulnerability to spoof attacks.

Anil K. Jain - One of the best experts on this subject based on the ideXlab platform.

  • hardening fingerprint fuzzy vault using password
    International Conference on Biometrics, 2007
    Co-Authors: Karthik Nandakumar, Abhishek Nagar, Anil K. Jain
    Abstract:

    Security of stored templates is a critical issue in biometric systems because biometric templates are non-revocable. Fuzzy vault is a cryptographic framework that enables secure template storage by binding the template with a uniformly random key. Though the fuzzy vault framework has proven security properties, it does not provide privacy-enhancing features such as revocability and protection against cross-matching across different biometric systems. Furthermore, non-uniform nature of biometric data can decrease the vault security. To overcome these limitations, we propose a scheme for hardening a fingerprint minutiae-based fuzzy vault using password. Benefits of the proposed password-based hardening technique include template revocability, prevention of cross-matching, enhanced vault security and a reduction in the False Accept Rate of the system without significantly affecting the False Reject Rate. Since the hardening scheme utilizes password only as an additional authentication factor (independent of the key used in the vault), the security provided by the fuzzy vault framework is not affected even when the password is compromised.

  • On-line signature verification
    Pattern Recognition, 2002
    Co-Authors: Anil K. Jain, Friederike D. Griess, Scott D. Connell
    Abstract:

    Abstract We describe a method for on-line handwritten signature verification. The signatures are acquired using a digitizing tablet which captures both dynamic and spatial information of the writing. After preprocessing the signature, several features are extracted. The authenticity of a writer is determined by comparing an input signature to a stored reference set (template) consisting of three signatures. The similarity between an input signature and the reference set is computed using string matching and the similarity value is compared to a threshold. Several approaches for obtaining the optimal threshold value from the reference set are investigated. The best result yields a False Reject Rate of 2.8% and a False accept Rate of 1.6%. Experiments on a database containing a total of 1232 signatures of 102 individuals show that writer-dependent thresholds yield better results than using a common threshold.

  • ICIP (2) - Automatic caption localization in compressed video
    Proceedings 1999 International Conference on Image Processing (Cat. 99CH36348), 1999
    Co-Authors: Yu Zhong, Hong-jiang Zhang, Anil K. Jain
    Abstract:

    We present a method to automatically locate captions in MPEG video. Caption text regions are segmented from the background using their distinguishing texture characteristics. This method first locates candidate text regions directly in the DCT compressed domain, and then reconstructs the candidate regions for further refinement in the spatial domain. Therefore, only a small amount of decoding is required. The proposed algorithm achieves about 4.0% False Reject Rate and less than 5.7% False positive Rate on a variety of MPEG compressed video containing more than 42,000 frames.

  • AVBPA - Identity Authentication Using Fingerprints
    Audio- and Video-based Biometric Person Authentication, 1997
    Co-Authors: Lin Hong, Anil K. Jain, Sharath Pankanti, Ruud M. Bolle
    Abstract:

    We describe the design and implementation of an automatic identity authentication system which uses fingerprint to establish the identity of an individual. An improved minutia extraction algorithm that is much faster and more accuRate than our earlier algorithm [12] has been implemented. An alignment-based elastic matching algorithm has been developed. This algorithm is capable of finding the correspondences between input minutia pattern and the stored template minutia pattern without resorting to exhaustive search and has the ability to adaptively compensate for the nonlinear deformations and inexact pose transformations between an input fingerprint and a template. The system has been tested on the MSU fingerprint database. A perfect authentication Rate can be achieved with a 15% False Reject Rate on this data set. Typically, a complete authentication procedure takes, on an average, about 1.4 seconds on a Sun ULTRA 1 workstation.