A New Order Selection Criterion for Autoregressive Models
A new criterion for choosing the model order required to fit an autoregressive model to measured ... more A new criterion for choosing the model order required to fit an autoregressive model to measured data is presented. The criterion operates on the eigenvalues resulting from the eigenvalue decomposi- tion of the model's covariance matrix and produces an estimate of the amplitude of.the errors, in the data as a byproduct. Since,Prony's method is equivalent to applying the covariance method to estimate the autoregressive model's parameters, this criterion can be viewed as a refinement of a previous criterion for F'rony's method developed by M. L. Van Blaricum. The criterion is demonstrated by way of numer- ical example. Abstract-Starting with the traditional spectrum estimation technique a new iterative nonparametric algorithm is developed. In each iteration of the algorithm, the effect of the previous spectral estimate is removed from the original signal by inverse filtering by a minimumyphase impulse response to obtain a residual signal whose spectral characteristics are then estimated to improve the previous estimate. The algorithmin each iteration decreases the bias of the resulting spectral estimate, without increasing its variance. Identification of the corresponding system min- imum phase finite duration impulse responseis inherent in the proposed procedure. The whole procedure can be cayied out with the numerical stability of FFT's, without solving any kind of equations. This algo- rithm is not suitable for very narrowband spectral shapes estimation from small amounts of data. Properties of the.proposed estimator are derived in detail. Experimental results show that the algorithm con- verges in a small number of iterations, and demonstrate that the algo- rithm provides very accurate spectral estimates compared with parametric methods based on ARMA models.
This paper discusses the advantages of score-level fusion between pattern and minutiae based fing... more This paper discusses the advantages of score-level fusion between pattern and minutiae based fingerprint verification algorithms in various operational scenarios. The different scenarios considered are sensor interoperability, environmental conditions and low quality enrollments. These are commonly encountered in real-life deployments of fingerprint-based biometric systems, specifically for large-scale distributed systems and physical access control. Moreover, the approach for jointly utilizing the conceptually different pattern and minutiae algorithms is based on various well-known scorelevel fusion techniques with single finger presentations. In contrast to previous studies on multi-matcher score-level fusion for fingerprint verification, where only moderate performance improvement were reported, the results presented here show significant performance gains. The two main contributing factors to these findings are that the two algorithms are conceptually different and the effects of the different operational scenarios. For the latter, improvement in accuracy due to fusion is even more significant in non-ideal and challenging operating conditions.
An efficient floating-point to fixed-point conversion process for biometric algorithm on DaVinci DSP architecture
Proceedings of SPIE, May 1, 2009
Today there is no direct path for the conversion of a floating-point algorithm implementation to ... more Today there is no direct path for the conversion of a floating-point algorithm implementation to an optimized fixed-point implementation. This paper proposes a novel and efficient methodology for Floating-point to Fixed-point Conversion (FFC) of biometric Fingerprint Algorithm ...
IEEE Transactions on Acoustics, Speech, and Signal Processing, Feb 1, 1984
Starting with the traditional spectrum estimation technique, a new iterative nonparametric algori... more Starting with the traditional spectrum estimation technique, a new iterative nonparametric algorithm is developed. In each iteration of the algorithm, the effect of the previous spectral estimate is removed from the original signal by inverse filtering by a minimumphase impulse response to obtain a residual signal whose spectral characteristics are then estimated to improve the previous estimate. The algorithm in each iteration decreases the bias of the resulting spectral estimate, without increasing its final variance. Identification of the corresponding system minimum-phase finite duration impulse response is inherent in the proposed procedure. The whole procedure can be canied out with the numerical stability of FFT's, without solving any kind of equations. This algorithm is not suitable for very narrowband spectral shapes estimation from small amounts of data. Properties of the proposed estimator are derived in detail. Experimental results show that the algorithm converges in a small number of iterations, and demonstrate that the algorithm provides very accurate spectral estimates compared with parametric methods based on ARMA models.
<title>An efficient floating-point to fixed-point conversion process for biometric algorithm on DaVinci DSP architecture</title>
Optics and Photonics in Global Homeland Security V and Biometric Technology for Human Identification VI, 2009
Today there is no direct path for the conversion of a floating-point algorithm implementation to ... more Today there is no direct path for the conversion of a floating-point algorithm implementation to an optimized fixed-point implementation. This paper proposes a novel and efficient methodology for Floating-point to Fixed-point Conversion (FFC) of biometric Fingerprint Algorithm Library (FAL) on fixed-point DaVinci processor. A general FFC research task is streamlined along smaller tasks which can be accomplished with lower effort and higher certainty. Formally specified in this paper is the optimization target in FFC, to preserve floating-point accuracy and to reduce execution time, while preserving the majority of algorithm code base. A comprehensive eight point strategy is formulated to achieve that target. Both local (focused on the most time consuming routines) and global optimization flow (to optimize across multiple routines) are used. Characteristic phases in the FFC activity are presented using data from employing the proposed FFC methodology to FAL, starting with target optimization specification, to speed optimization breakthroughs, finalized with validation of FAL accuracy after the execution time optimization. FAL implementation resulted in biometric verification time reduction for over a factor of 5, with negligible impact on accuracy. Any algorithm developer facing the task of implementing his floating-point algorithm on DaVinci DSP is expected to benefit from this presentation.
This paper discusses the advantages of score-level fusion between pattern and minutiae based fing... more This paper discusses the advantages of score-level fusion between pattern and minutiae based fingerprint verification algorithms in various operational scenarios. The different scenarios considered are sensor interoperability, environmental conditions and low quality enrollments. These are commonly encountered in real-life deployments of fingerprint-based biometric systems, specifically for large-scale distributed systems and physical access control. Moreover, the approach for jointly utilizing the conceptually different pattern and minutiae algorithms is based on various well-known scorelevel fusion techniques with single finger presentations. In contrast to previous studies on multi-matcher score-level fusion for fingerprint verification, where only moderate performance improvement were reported, the results presented here show significant performance gains. The two main contributing factors to these findings are that the two algorithms are conceptually different and the effects of the different operational scenarios. For the latter, improvement in accuracy due to fusion is even more significant in non-ideal and challenging operating conditions.
A new class of two-dimensional windows for bispectrum estimation
Signal Processing, 1994
A new class of bispectral windows is proposed, starting from the symmetry characteristics of thir... more A new class of bispectral windows is proposed, starting from the symmetry characteristics of third-order moments. Desirable properties and expressions of bispectral windows are summarized to support the adoption of minimum mean square error factor of the bispectral estimate as ...
Verification Speed in Fingerprint-Based Biometric Systems
International Journal of Image and Graphics, 2003
Issues with verification speed improvement to fingerprint systems are investigated in this paper.... more Issues with verification speed improvement to fingerprint systems are investigated in this paper. First, the impact of verification speed on the overall system performance is highlighted. Then a rather general speed enhancement procedure for algorithms is proposed, consisting of two or more pattern matching stages for refining the value of the discrimination function. Special attention is paid to the algorithm supporting the verification procedure used to determine the two verification thresholds. It is shown, and supported by numerical examples, that significant verification speed improvement can be achieved without sacrificing the system accuracy.
IEEE Transactions on Acoustics, Speech, and Signal Processing, 1984
Starting with the traditional spectrum estimation technique, a new iterative nonparametric algori... more Starting with the traditional spectrum estimation technique, a new iterative nonparametric algorithm is developed. In each iteration of the algorithm, the effect of the previous spectral estimate is removed from the original signal by inverse filtering by a minimumphase impulse response to obtain a residual signal whose spectral characteristics are then estimated to improve the previous estimate. The algorithm in each iteration decreases the bias of the resulting spectral estimate, without increasing its final variance. Identification of the corresponding system minimum-phase finite duration impulse response is inherent in the proposed procedure. The whole procedure can be canied out with the numerical stability of FFT's, without solving any kind of equations. This algorithm is not suitable for very narrowband spectral shapes estimation from small amounts of data. Properties of the proposed estimator are derived in detail. Experimental results show that the algorithm converges in a small number of iterations, and demonstrate that the algorithm provides very accurate spectral estimates compared with parametric methods based on ARMA models.
Uploads
Papers by Ira Konvalinka