Po-Hsiang Lai

ECG biometrics: A robust short-time frequency analysis

In this paper, we present the results of an analysis of the electrocardiogram (ECG) as a biométrie using a novel short-time frequency method with robust feature selection. Our proposed method incorporates heartbeats from multiple days and fuses information. Single lead ECG signals from a comparatively large sample of 269 subjects that were sampled from the general population were collected on three separate occasions over a seven-month period. We studied the impact of long-term variability, health status, data fusion, the number of training and testing heartbeats, and database size on ECG biométrie performance. The proposed method achieves 5.58% equal error rate (EER) in verification, 76.9% accuracy in rank-1 recognition, and 93.5% accuracy in rank-15 recognition when training and testing heartbeats are from different days. If training and testing heartbeats are collected on the same day, we achieve 0.37% EER and 99% recognition accuracy for decisions based on a single heartbeat.

Laser Doppler Vibrometry Measures of Physiological Function: Evaluation of Biometric Capabilities

A novel approach for remotely sensing mechanical cardiovascular activity for use as a biometric marker is proposed. Laser Doppler Vibrometry (LDV) is employed to sense vibrations on the surface of the skin above the carotid artery related to arterial wall movements associated with the central blood pressure pulse. Carotid LDV signals are recorded using noncontact methods and the resulting unobtrusiveness is a major benefit of this technique. Several recognition methods are proposed that use the temporal and/or spectral information in the signal to assess biometric performance both on an intrasession basis, and on an intersession basis where LDV measurements were acquired from the same subjects after delays ranging from one week to six months. A waveform decomposition method that utilizes principal component analysis is used to model the signal in the time domain. Authentication testing for this approach produces an equal-error rate of 0.5% for intrasession testing. However, performance degrades substantially for intersession testing, requiring a more robust approach to modeling. Improved performance is obtained using techniques based on time-frequency decomposition, incorporating a method for extracting informative components. Biometric fusion methods including data fusion and information fusion are applied to train models using data from multiple sessions. As currently implemented, this approach yields an intersession equal-error rate of 6.3%.

Hidden State Models for Noncontact Measurements of the Carotid Pulse Using a Laser Doppler Vibrometer

The method of laser Doppler vibrometry (LDV) is used to sense movements of the skin overlying the carotid artery. When pointed at the skin overlying the carotid artery, the mechanical movements of the skin disclose physiological activity relating to the blood pressure pulse over the cardiac cycle. In this paper, signal modeling is addressed, with close attention to the underlying physiology. Segments of the LDV signal corresponding to single heartbeats, called LDV pulses, are extracted. Hidden Markov models (HMMs) are used to capture the dynamics of the LDV pulses from beat to beat based on pulse morphology; under resting conditions these dynamics are primarily due to respiration-related effects. LDV pulses are classified according to state, by computing the optimal state path through the data using trained HMMs. HMM state dynamics are examined within the context of respiratory effort using strain gauges placed around the abdomen. This study presented here provides a graphical model approach to modeling the dependence of the LDV pulse on latent states.

Biometrics with physical exercise using Laser Doppler Vibrometry measurements of the carotid pulse

Understanding the variability of the cardiacrelated signals caused by physical exercise is an interesting and important problem. To our knowledge, there is no paper evaluating the biometric consistency of the cardiovascular based signals during the physical exercise, or the extent to which the signals can recover after that. A novel method of remotely sensing mechanical activity related to the carotid pulse with Laser Doppler Vibrometry (LDV) has been developed. Encouraging results are obtained on the evaluation of the LDV cardiovascular signal as a biometric marker. A new protocol is set up to produce changes in heart rate by physical exercise. Spectral based approaches are applied following the success in general biometric authentication. An equal error rate of 2.8% for inter-state tests indicates that the LDV pulse signal is quite stable even after moderate physical exercise. The performance degrades during exercise, especially when the heart rate reaches 55% of the age-adjusted theoretical maximum heart rate. When the test individuals start resting, the performance improves as the heart rate recovered within seconds. We can say that the short-term variability caused by heart rate fluctuations and respiration changes recover with enough stability in a short time for biometric consistency.

A robust feature selection method for noncontact biometrics based on Laser Doppler Vibrometry

We propose a new biometric approach based on cardiovascular signals recorded using laser Doppler vibrometry (LDV) with a robust feature selection method. A novel feature selection method provides robustness against physiological variability of a given individual. LDV signals were collected from 191 individuals under controlled conditions during three sessions, each at intervals of one week to six months. The methods described here are based on a time-frequency decomposition of the LDV signal in which the log-power of the decomposition values are used as features. In identity verification tasks, equal error rates in the single digits can be achieved with testing periods as short as 4 s.

Hidden state dynamics in laser Doppler vibrometery measurements of the carotid pulse under resting conditions

A laser Doppler vibrometer (LDV) is used to sense movements of the skin overlying the carotid artery. Fluctuations in carotid artery diameter due to variations in the underlying blood pressure are sensed at the surface of the skin. Portions of the LDV signal corresponding to single heartbeats, called the LDV pulses, are extracted. This paper introduces the use of hidden Markov models (HMMs) to model the dynamics of the LDV pulse from beat to beat based on pulse morphology, which under resting conditions are primarily due to breathing effects. LDV pulses are classified according to state, by computing the optimal state path through the data using trained HMMs. HMM state dynamics are compared to simultaneous recordings of strain gauges placed on the abdomen. The work presented here provides a robust statistical approach to modeling the dependence of the LDV pulse on latent states.

Minimum description length and clustering with exemplars

We propose an information-theoretic clustering framework for density-based clustering and similarity or distance-based clustering with objective functions of clustering performance derived from stochastic complexity and minimum description length (MDL) arguments. Under this framework, the number of clusters and parameters can be determined in a principled way without prior knowledge from users. We show that similarity-based clustering can be viewed as combinatorial optimization on graphs. We propose two clustering algorithms, one of which relies on a minimum arborescence tree algorithm which returns optimal clustering under the proposed MDL objective function for similarity-based clustering. We demonstrate clustering performance on synthetic data.

Pattern Recognition System Design with Linear Encoding for Discrete Patterns

Pattern recognition systems based on compressed patterns and compressed sensor measurements can be designed using low-density matrices. We examine truncation encoding where a subset of the patterns and measurements are stored perfrectly while the rest is discarded. We also examine the use of LDPC parity check matrices for compressing measurements and patterns. We show how more general ensembles of good linear codes can be used as the basis for pattern recognition system design, yielding system design strategies for more general noise models.

Pattern recognition system design based on LDPC matrices

Pattern recognition systems may be designed to recognize an exponentially large number of objects from potentially noisy measurements. We propose a design based on storing compressed representations of binary patterns corresponding to objects of interest. Sensor measurements are similarly compressed and recognition proceeds by comparing the compressed sensor measurements to the compressed representations of the objects. Parity check matrices corresponding to low density parity check codes are used for the compression. This design yields an ensemble of systems such that the probability of error goes to zero as the length of the patterns grows

MDL hierarchical clustering with incomplete data

The goal of stemmatology is to reconstruct a family tree of different variants of a text resulting from imperfect copying, which is a crucial part of textual criticism. In reality, historians often have incomplete data because some variants are not yet discovered and there are missing portions in available variants due to physical damage. Stemmatology is similar to molecular phylogenetics where biologists aim to reconstruct the evolutionary history of species based on genetic or protein sequences. Adoption of phylogenetics methods has lead to encouraging results in automatic stemmatology. We discuss and demonstrate the potential application of minimum description length (MDL) concepts to stemmatology. Our method is applied to a realistic dataset and outperforms major existing methods.