Ainara Garde

Audio, video and multimodal person identification in a smart room

In this paper, we address the modality integration issue on the example of a smart room environment aiming at enabling person identification by combining acoustic features and 2D face images. First we introduce the monomodal audio and video identification techniques and then we present the use of combined input speech and face images for person identification. The various sensory modalities, speech and faces, are processed both individually and jointly. Its shown that the multimodal approach results in improved performance in the identification of the participants.

Analysis of Respiratory Flow Signals in Chronic Heart Failure Patients with Periodic Breathing

In patients with chronic heart failure (CHF), oscillatory breathing pattern predicts poor prognosis. This work proposes a method to identify the respiratory pattern to determine periodic breathing (PB), Cheyne-Stokes respiration (CSR) and non-periodic respiratory patterns (nPB) through the respiratory flow signal. 26 patients are studied, classified in G1 (PB), G2 (CSR) and G3 (nPB). The flow signal is filtered and normalized, to obtain the positive envelope that describes the respiratory pattern. With this new signal some features are extracted through its power spectral density (PSD). An adaptive feature selection algorithm is applied before the linear and non linear classification applying Leave-one-out cross-validation technique. The result obtained with linear classification was 93% using the relation between total energy and frequency interval (ll), peak amplitude (ampp), peak frequency (fp), and the highest slope of the positive envelopes PSD (Slopemax). And the best result was obtained with non linear technique, with 100% correctly classified patients, using only two parameters, fp and Slopemax.

Characterization of periodic and non-periodic breathing pattern in chronic heart failure patients

Periodic breathing (PB) has a high prevalence in chronic heart failure (CHF) patients with mild to moderate symptoms and poor ventricular function. This work proposes the analysis and characterization of the respiratory pattern to identify periodic breathing pattern (PB) and non-periodic breathing pattern (nPB) through the respiratory flow signal. The respiratory pattern analysis is based on the extraction and the study of the flow envelope signal. The flow envelope signal is modelled by an autoregressive model (AR) whose coefficients would characterize the respiratory pattern of each group. The goodness of the characterization is evaluated through a linear and non linear classifier applied to the AR coefficients. An adaptive feature selection is used before the linear and non linear classification, employing leave-one-out cross validation technique. With linear classification the percentage of well classified patients (8 PB and 18 nPB patients) is 84.6% using the statistically significant coefficients whereas with non linear classification, the percentage of well classified patients increase to more than 92% applying the best subset of coefficients extracted by a forward selection algorithm.

On the use of genuine-impostor statistical information for score fusion in multimodal biometrics/sur l’usage de l’information statistioue client-imposteur pour la fusion des scores en biométrie multimodale

Matching score level fusion techniques in multimodal person verification conventionally use global score statistics in the normalization and fusion stages. In this paper, novel normalization and fusion methods are presented to take advantage of the separate statistics of the monomodal scores in order to reduce the genuine and impostor pdf lobe overlapping and improve the verification rate. Joint mean normalization is an affine transformation that normalizes the mean of the monomodal biometrics scores separately for the genuine and impostor individuals. Histogram equalization is used to align the statistical distribution of the monomodal scores and make the whole separate statistics comparable. The presented weighting fusion methods have been designed to minimize the variances of the separate multimodal statistics and reduce the multimodal pdf lobe overlapping. The results obtained in speech and face scores fusion upon polycost and xm2vts databases show that the proposed techniques provide better results than the conventional methods.RésuméLes techniques de fusion au niveau des degrés de pertinence dans la vérification multimodale de personnes utilisent conventionnellement des statistiques globales de perticence pour les étapes de normalisation et de fusion. Dans le présent article, de nouvelles méthodes de normalisation et de fusion sont présentées pour profiter des statistiques séparées des pertinences monomodales en vue de réduire la superposition des densités de probabilité de client et d’imposteur et d’améliorer le taux de vérification. La normalisation conjointe de la moyenne est une transformation affine qui normalise la moyenne des qualifications biométriques monomodales séparément pour les individus client et imposteur. L’égalisation de l’histogramme est utilisée pour aligner la distribution statistique des pertinences monomodales et peut rendre comparables les statistiques complètes séparées. Les présentes méthodes de fusion avec pondération on été conçues de façon à minimiser les variances des statistiques multimodales séparées et réduire la superposition des densités de probabilité multimodales. Les résultats obtenus dans la fusion de pertinences pour voix et visage avec les bases de données polycost et xm2vts démontrent que la normalisation proposée et les techniques de fusion produisent de meilleurs résultats que les méthodes conventionnelles.

Time-varying respiratory pattern characterization in chronic heart failure patients and healthy subjects

Patients with chronic heart failure (CHF) with periodic breathing (PB) and Cheyne–Stokes respiration (CSR) tend to exhibit higher mortality and poor prognosis. This study proposes the characterization of respiratory patterns in CHF patients and healthy subjects using the envelope of the respiratory flow signal, and autoregressive (AR) time–frequency analysis. In time-varying respiratory patterns, the statistical distribution of the AR coefficients, pole locations, and the spectral parameters that characterize the discriminant band are evaluated to identify typical breathing patterns. In order to evaluate the accuracy of this characterization, a feature selection process followed by linear discriminant analysis is applied. 26 CHF patients (8 patients with PB pattern and 18 with non-periodic breathing pattern (nPB)) are studied. The results show an accuracy of 83.9% with the mean of the main pole magnitude and the mean of the total power, when classifying CHF patients versus healthy subjects, and 83.3% for nPB versus healthy subjects. The best result when classifying CHF patients into PB and nPB was an accuracy of 88.9%, using the coefficient of variation of the first AR coefficient and the mean of the total power.

Periodic breathing during ascent to extreme altitude quantified by spectral analysis of the respiratory volume signal

High altitude periodic breathing (PB) shares some common pathophysiologic aspects with sleep apnea, Cheyne-Stokes respiration and PB in heart failure patients. Methods that allow quantifying instabilities of respiratory control provide valuable insights in physiologic mechanisms and help to identify therapeutic targets. Under the hypothesis that high altitude PB appears even during physical activity and can be identified in comparison to visual analysis in conditions of low SNR, this study aims to identify PB by characterizing the respiratory pattern through the respiratory volume signal. A number of spectral parameters are extracted from the power spectral density (PSD) of the volume signal, derived from respiratory inductive plethysmography and evaluated through a linear discriminant analysis. A dataset of 34 healthy mountaineers ascending to Mt. Muztagh Ata, China (7,546 m) visually labeled as PB and non periodic breathing (nPB) is analyzed. All climbing periods within all the ascents are considered (total climbing periods: 371 nPB and 40 PB). The best crossvalidated result classifying PB and nPB is obtained with Pm (power of the modulation frequency band) and R (ratio between modulation and respiration power) with an accuracy of 80.3% and area under the receiver operating characteristic curve of 84.5%. Comparing the subjects from 1st and 2nd ascents (at the same altitudes but the latter more acclimatized) the effect of acclimatization is evaluated. SaO2 and periodic breathing cycles significantly increased with acclimatization (p-value <; 0.05). Higher Pm and higher respiratory frequencies are observed at lower SaO2, through a significant negative correlation (p-value <; 0.01). Higher Pm is observed at climbing periods visually labeled as PB with >; 5 periodic breathing cycles through a significant positive correlation (p-value <; 0.01). Our data demonstrate that quantification of the respiratory volume signal using spectral analysis is suitable to identify effects of hypobaric hypoxia on control of breathing.

Correntropy-based nonlinearity test applied to patients with chronic heart failure

In this study we propose the correntropy function as a discriminative measure for detecting nonlinearities in the respiratory pattern of chronic heart failure (CHF) patients with periodic or nonperiodic breathing pattern (PB or nPB, respectively). The complexity seems to be reduced in CHF patients with higher risk level. Correntropy reflects information on both, statistical distribution and temporal structure of the underlying dataset. It is a suitable measure due to its capability to preserve nonlinear information. The null hypothesis considered is that the analyzed data is generated by a Gaussian linear stochastic process. Correntropy is used in a statistical test to reject the null hypothesis through surrogate data methods. Various parameters, derived from the correntropy and correntropy spectral density (CSD) to characterize the respiratory pattern, presented no significant differences when extracted from the iteratively refined amplitude adjusted Fourier transform (IAAFT) surrogate data. The ratio between the powers in the modulation and respiratory frequency bands R was significantly different in nPB patients, but not in PB patients, which reflects a higher presence of nonlinearities in nPB patients than in PB patients.

Correntropy-based analysis of respiratory patterns in patients with chronic heart failure

A correntropy-based technique is proposed for the analysis and characterization of respiratory flow signals in chronic heart failure (CHF) patients with both periodic and nonperiodic breathing (PB and nPB), and healthy subjects. Correntropy is a novel similarity measure which provides information on temporal structure and statistical distribution simultaneously. Its properties lend itself to the definition of the correntropy spectral density (CSD). An interesting result from CSD-based spectral analysis is that both the respiratory frequency and modulation frequency can be detected at their original positions in the spectrum without prior demodulation of the flow signal. The respiratory pattern is characterized by a number of spectral parameters extracted from the respiratory and modulation frequency bands. The results show that the power of the modulation frequency band offers excellent performance when classifying CHF patients versus healthy subjects, with an accuracy of 95.3%, and nPB patients versus healthy subjects with 90.7%. The ratio between the power in the modulation and respiration frequency bands provides the best results classifying CHF patients into PB and nPB, with an accuracy of 88.9%.

Person Verification by Fusion of Prosodic, Voice Spectral and Facial Parameters.

This work has been partially supported by the European Union (under CHIL IST-2002-506909 and BIOSEC IST-2002-001766) and by the Spanish Government (under ACESCA project TIN2005-08852 and grant AP2003-3598).

Analysis of the respiratory flow cycle morphology in chronic heart failure patients applying principal components analysis

The study of flow cycle morphology provides new information about the breathing pattern. This study proposes the characterization of cycle morphology in chronic heart failure patients (CHF) patients, with periodic (PB) and non-periodic breathing (nPB) patterns, and healthy subjects. Principal component analysis is applied to extract a respiratory cycle model for each time segment defined by a 30-s moving window. To characterize morphology of the model waveform, a number of parameters are extracted whose significance is evaluated in terms of the following three classification problems: CHF patients with either PB or nPB, CHF patients versus healthy subjects, and nPB patients versus healthy subjects. 26 CHF patients (8 with PB and 18 with non-periodic breathing pattern (nPB)) and 35 healthy subjects are studied. The results show that a respiratory cycle compressed in time characterizes PB patients, i.e., shorter inspiratory and expiratory periods, and higher dispersion of the maximum inspiratory and expiratory flow value (accuracy of 87%). The maximal expiratory flow instant occurs earlier in CHF patients than in healthy subjects (accuracy of 87%), with a steeper slope between inspiration and expiration. It is also found that the standard deviation of the expiratory period, evaluated for each subject, is much lower in CHF patients than in healthy subjects. The maximal expiratory flow instant occurs earlier (accuracy of 84%) in nPB patients, when comparing subjects with similar respiratory pattern like nPB patients and healthy subjects.