Property:Abstract
From ISLAB/CAISR
Jump to navigationJump to searchThis is a property of type Text.
P
<p>We present a new system for biometric recognition using periocular images based on retinotopic sampling grids and Gabor analysis of the local power spectrum. A number of aspects are studied, including: 1) grid adaptation to dimensions of the target eye vs. grids of constant size, 2) comparison between circular- and rectangular-shaped grids, 3) use of Gabor magnitude vs. phase vectors for recognition, 4) rotation compensation between query and test images, and 5) comparison with an iris machine expert. Results show that our system achieves competitive verification rates compared with other periocular recognition approaches. We also show that top verification rates can be obtained without rotation compensation, thus allowing to remove this step for computational efficiency. Also, the performance is not affected substantially if we use a grid of fixed dimensions, or it is even better in certain situations, avoiding the need of accurate detection of the iris region. © 2012 Springer-Verlag.</p> +
<p>We present a new system for biometric recognition using periocular images. The feature extraction method employed describes neighborhoods around keypoints by projection onto harmonic functions which estimates the presence of a series of various symmetric curve families around such keypoints. The iso-curves of such functions are highly symmetric w.r.t. the keypoints and the estimated coefficients have well defined geometric interpretations. The descriptors used are referred to as Symmetry Assessment by Feature Expansion (SAFE). Extraction is done across a set of discrete points of the image, uniformly distributed in a rectangular-shaped grid positioned in the eye center. Experiments are done with two databases of iris data, one acquired with a close-up iris camera, and another in visible light with a webcam. The two databases have been annotated manually, meaning that the radius and center of the pupil and sclera circles are available, which are used as input for the experiments. Results show that this new system has a performance comparable with other periocular recognition approaches. We particularly carry out comparative experiments with another periocular system based on Gabor features extracted from the same set of grid points, with the fusion of the two systems resulting in an improved performance. We also evaluate an iris texture matcher, providing fusion results with the periocular systems as well.</p> +
<p>Periocular biometrics has been established as an independent modality due to concerns on the performance of iris or face systems in uncontrolled conditions. Periocular refers to the facial region in the eye vicinity, including eyelids, lashes and eyebrows. It is available over a wide range of acquisition distances, representing a tradeoff between the whole face (which can be occluded at close distances) and the iris texture (which do not have enough resolution at long distances). Since the periocular region appears in face or iris images, it can be used also in conjunction with these modalities. Features extracted from the periocular region have been also used successfully for gender classification and ethnicity classification, and to study the impact of gender transformation or plastic surgery in the recognition performance. This paper presents a review of the state of the art in periocular biometric research, providing an insight of the most relevant issues and giving a thorough coverage of the existing literature. Future research trends are also briefly discussed.</p> +
<p>This paper describes a new motion based feature extraction technique for speaker recognition using orientation estimation in 2D manifolds. The motion is estimated by computing the components of the structure tensor from which normal flows are extracted. By projecting the 3D spatiotemporal data to 2-D planes we obtain projection coefficients which we use to evaluate the 3-D orientations of brightness patterns in TV like 2D image sequences. This corresponds to the solutions of simple matrix eigenvalue problems in 2D, affording increased computational efficiency. An implementation based on joint lip movements and speech is presented along with experiments which confirm the theory, exhibiting a recognition rate of 98% on the publicly available XM2VTS database.</p> +
<p>Detection and recognition of objects representing the Prorocentrum minimum (P. minimum) species in phytoplankton images is the main objective of the article. The species is known to cause harmful blooms in many estuarine and coastal environments. A new technique, combining phase congruency-based detection of circular objects in images, stochastic optimization-based object contour determination, and SVM- as well as random forest (RF)-based classification of objects was developed to solve the task. A set of various features including a subset of new features computed from phase congruency preprocessed images was used to characterize extracted objects. The developed algorithms were tested using 114 images of 1280×960 pixels. There were 2088 P. minimum cells in the images in total. The algorithms were able to detect 93.25% of objects representing P. minimum cells and correctly classified 94.9% of all detected objects. The feature set used has shown considerable tolerance to out-of-focus distortions. The obtained results are rather encouraging and will be used to develop an automated system for obtaining abundance estimates of the species. © 2011 Elsevier Ltd All rights reserved.</p> +
Physiologically inspired signal preprocessing for auditory prostheses : Insights from the electro-motility of the OHC +
<p>We designed a non-linear functional model of the outer hair cell (OHC) functioning in the filtering system of the cochlea and then isolated from it two second-order structures, one employing the mechanism of the somatic motility and the other the hair bundle motion of the OHC. The investigation of these circuits showed that the main mechanism increasing the sensitivity and frequency selectivity of the filtering system is the somatic motility. The mechanism of the active hair bundle motion appeared less suitable for realization of the band-pass filtering structures due to the dependence of the sensitivity, natural frequency and selectivity on the signal intensity. We combined three second-order filtering structures employing the mechanism of the somatic motility and the lateral inhibition to form a parallel-type filtering channel of the sixth order with the frequency characteristics of the Butterworth-type and Gaussian-type. The investigation of these channels showed that the Gaussian-type channel has the advantage over the Butterworth-type channel. It is more suitable for realization of a filter bank with common lateral circuits and has less distorted frequency characteristic in the nonlinear mode.</p> +
Pitfalls of Affective Computing : How can the automatic visual communication of emotions lead to harm, and what can be done to mitigate such risks? +
<p>What would happen in a world where people could "see'' others' hidden emotions directly through some visualizing technology Would lies become uncommon and would we understand each other better Or to the contrary, would such forced honesty make it impossible for a society to exist The science fiction television show Black Mirror has exposed a number of darker scenarios in which such futuristic technologies, by blurring the lines of what is private and what is not, could also catalyze suffering. Thus, the current paper first turns an eye towards identifying some potential pitfalls in emotion visualization which could lead to psychological or physical harm, miscommunication, and disempowerment. Then, some countermeasures are proposed and discussed--including some level of control over what is visualized and provision of suitably rich emotional information comprising intentions--toward facilitating a future in which emotion visualization could contribute toward people's well-being. The scenarios presented here are not limited to web technologies, since one typically thinks about emotion recognition primarily in the context of direct contact. However, as interfaces develop beyond today's keyboard and monitor, more information becomes available also at a distance--for example, speech-to-text software could evolve to annotate any dictated text with a speaker's emotional state.</p> +
Pitfalls of medication adherence approximation through EHR and pharmacy records : Definitions, data and computation +
<p>Background and purpose: Patients’ adherence to medication is a complex, multidimensional phenomenon. Dispensation data and electronic health records are used to approximate medication-taking through refill adherence. In-depth discussions on the adverse effects of data quality and computational differences are rare. The purpose of this article is to evaluate the impact of common pitfalls when computing medication adherence using electronic health records.</p><p>Procedures: We point out common pitfalls associated with the data and operationalization of adherence measures. We provide operational definitions of refill adherence and conduct experiments to determine the effect of the pitfalls on adherence estimations. We performed statistical significance testing on the impact of common pitfalls using a baseline scenario as reference.</p><p>Findings: Slight changes in definition can significantly skew refill adherence estimates. Pickup patterns cause significant disagreement between measures and the commonly used proportion of days covered. Common data related issues had a small but statistically significant (p < 0.05) impact on population-level and significant effect on individual cases.</p><p>Conclusion: Data-related issues encountered in real-world administrative databases, which affect various operational definitions of refill adherence differently, can significantly skew refill adherence values, leading to false conclusions about adherence, particularly when estimating adherence for individuals. © 2020 The Authors. Published by Elsevier B.V. </p> +
<p>In this paper, we present a hierarchical algorithm to register a partial fingerprint against a full fingerprint using only the orientation fields. In the first level, we shortlist possible locations for registering the partial fingerprint in the full fingerprint using a normalized correlation measure, taking various rotations into account. As a second level, on those candidate locations, we calculate three other similarity measures. We then perform score fusion for all the estimated similarity scores to locate the final registration. By registering a partial fingerprint against a full fingerprint, we can reduce the search space of the minutiae set in the full fingerprint, thereby improving the result of partial fingerprint identification, particularly for poor quality latent fingerprints. We report the rank identification improvements of two minutiae-based automated identification systems on the NIST-SD27 database when we use our hierarchical registration as a pre-alignment.</p> +
<p>Comparing a latent fingerprint minutiae set against a ten print fingerprint minutiae set using an automated fingerprint identification system is a challenging problem. This is mainly because latent fingerprints obtained from crime scenes are mostly partial fingerprints, and most automated systems expect approximately the same number of minutiae between query and the reference fingerprint under comparison for good performance. In this work, we propose a methodology to reduce the minutiae set of ten print with respect to that of query latent minutiae set by registering the orientation field of latent fingerprint with the ten print orientation field. By reducing the search space of minutiae from the ten print, we can improve the performance of automated identification systems for latent fingerprints. We report the performance of our registration algorithm on the NIST-SD27 database as well as the improvement in the Rank Identification accuracy of a standard minutiae-based automated system. © 2014 IEEE.</p> +
<p>In this study, the authors present a hierarchical algorithm to register a partial fingerprint against a full fingerprint using only the orientation fields. In the first level, they shortlist possible locations for registering the partial fingerprint in the full fingerprint using a normalised correlation measure, taking various rotations into account. As a second level, on those candidate locations, they calculate three other similarity measures. They then perform score fusion for all the estimated similarity scores to locate the final registration. By registering a partial fingerprint against a full fingerprint, they can reduce the search space of the minutiae set in the full fingerprint, thereby improving the result of partial fingerprint identification, particularly for poor quality latent fingerprints. They report the rank identification improvements of two minutiae-based automated fingerprint identification systems on the National Institute of Standards and Technology (NIST)-Special Database 27 database when they use the authors hierarchical registration as a pre-alignment. © The Institution of Engineering and Technology 2015.</p> +
<p>As human drivers, we instinctively employ our understanding of other road users' behaviour for enhanced efficiency of our drive and safety of the traffic. In recent years, different aspects of assisted and autonomous driving have gotten a lot of attention from the research and industrial community, including the aspects of behaviour modelling and prediction of future state. In this paper, we address the problem of modelling and predicting agent behaviour and state in a roundabout traffic scenario. We present three ways of modelling traffic in a roundabout based on: (i) the roundabout geometry; (ii) mean path taken by vehicles inside the roundabout; and (iii) a set of reference trajectories traversed by vehicles inside the roundabout. The roundabout models are compared in terms of exit-direction classification and state (i.e., position inside the roundabout) prediction of query vehicles inside the roundabout. The exit-direction classification and state prediction are based on a particle-filter classifier algorithm. The results show that the roundabout model based on set of reference trajectories is better suited for both the exit-direction and state prediction.</p> +
<p>We introduce an LSTM-based method for predicting compressor failures using aggregated sensory data, and evaluate it using historical information from over 1000 heavy duty vehicles during 2015 and 2016. The goal is to proactively identify trucks that will require maintenance in the near future, so that component replacement can be scheduled before the failure happens, translating into improved uptime. The problem is formulated as a classification task of whether a compressor failure will happen within the specified prediction horizon. A recurrent neural network using Long Short-Term Memory (LSTM) architecture is employed as the prediction model, and compared against Random Forest (RF), the solution used in industrial deployment at the moment. Experimental results show that while Random Forest slightly outperforms LSTM in terms of AUC score, the predictions of LSTM stay significantly more stable over time, showing a consistent trend from healthy to faulty class. Additionally, LSTM is also better at detecting the switch from faulty class to the healthy one after a repair. We demonstrate that this stability is important for making repair decisions, especially in questionable cases, and therefore LSTM model is likely to lead to better results in practice.</p> +
<p>Modern vehicles have increasing amounts of data streaming continuously on-board their controller area networks. These data are primarily used for controlling the vehicle and for feedback to the driver, but they can also be exploited to detect faults and predict failures. The traditional diagnostics paradigm, which relies heavily on human expert knowledge, scales poorly with the increasing amounts of data generated by highly digitised systems. The next generation of equipment monitoring and maintenance prediction solutions will therefore require a different approach, where systems can build up knowledge (semi-)autonomously and learn over the lifetime of the equipment.</p><p>A key feature in such systems is the ability to capture and encode characteristics of signals, or groups of signals, on-board vehicles using different models. Methods that do this robustly and reliably can be used to describe and compare the operation of the vehicle to previous time periods or to other similar vehicles. In this paper two models for doing this, for a single signal, are presented and compared on a case of on-road failures caused by air compressor faults in city buses. One approach is based on histograms and the other is based on echo state networks. It is shown that both methods are sensitive to the expected changes in the signal's characteristics and work well on simulated data. However, the histogram model, despite being simpler, handles the deviations in real data better than the echo state network.</p> +
Predicting Blood Lactate Concentration and Oxygen Uptake from sEMG Data during Fatiguing Cycling Exercise +
<p>This article presents a study of the relationship between electromyographic (EMG) signals from vastus lateralis, rectus femoris, biceps femoris and semitendinosus muscles, collected during fatiguing cycling exercises, and other physiological measurements, such as blood lactate concentration and oxygen consumption. In contrast to the usual practice of picking one particular characteristic of the signal, e.g., the median or mean frequency, multiple variables were used to obtain a thorough characterization of EMG signals in the spectral domain. Based on these variables, linear and non-linear (random forest) models were built to predict blood lactate concentration and oxygen consumption. The results showed that mean and median frequencies are sub-optimal choices for predicting these physiological quantities in dynamic exercises, as they did not exhibit significant changes over the course of our protocol and only weakly correlated with blood lactate concentration or oxygen uptake. Instead, the root mean square of the original signal and backward difference, as well as parameters describing the tails of the EMG power distribution were the most important variables for these models. Coefficients of determination ranging from R<sup>2</sup> = 0:77 to R<sup>2</sup> = 0:98 (for blood lactate) and from R<sup>2</sup> = 0:81 to R<sup>2</sup> = 0:97 (for oxygen uptake) were obtained when using random forest regressors.</p> +
<p>The rising complexity in healthcare, exacerbated by an ageing population, results in ineffective decision-making leading to detrimental effects on care quality and escalates care costs. Consequently, there is a need for smart decision support systems that can empower clinician's to make better informed care decisions. Decisions, which are not only based on general clinical knowledge and personal experience, but also rest on personalised and precise insights about future patient outcomes. A promising approach is to leverage the ongoing digitization of healthcare that generates unprecedented amounts of clinical data stored in Electronic Health Records (EHRs) and couple it with modern Machine Learning (ML) toolset for clinical decision support, and simultaneously, expand the evidence base of medicine. As promising as it sounds, assimilating complete clinical data that provides a rich perspective of the patient's health state comes with a multitude of data-science challenges that impede efficient learning of ML models. This thesis primarily focuses on learning comprehensive patient representations from EHRs. The key challenges of heterogeneity and temporality in EHR data are addressed using human-derived features appended to contextual embeddings of clinical concepts and Long-Short-Term-Memory networks, respectively. The developed models are empirically evaluated in the context of predicting adverse clinical outcomes such as mortality or hospital readmissions. We also present evidence that, surprisingly, different ML models primarily designed for non-EHR analysis (like language processing and time-series prediction) can be combined and adapted into a single framework to efficiently represent EHR data and predict patient outcomes.</p> +
Predicting physiological parameters in fatiguing bicycling exercises using muscle activation timing +
<p>This article is concerned with a novel technique for prediction of blood lactate concentration level and oxygen uptake rate from multi-channel surface electromyography (sEMG) signals. The approach is built on predictive models exploiting a set of novel time-domain variables computed from sEMG signals. Signals from three muscles of each leg, namely, vastus lateralis, rectus femoris, and semitendinosus were used in this study. The feature set includes parameters reflecting asymmetry between legs, phase shifts between activation of different muscles, active time percentages, and sEMG amplitude. Prediction ability of both linear and non-linear (random forests-based) models was explored. The random forests models showed very good prediction accuracy and attained the coefficient of determination R<sup>2</sup> = 0.962 for lactate concentration level and R<sup>2</sup> = 0.980 for oxygen uptake rate. The linear models showed lower prediction accuracy. Comparable results were obtained also when sEMG amplitude data were removed from the training sets. A feature elimination algorithm allowed to build accurate random forests (R<sup>2</sup> > 0.9) using just six (lactate concentration level) or four (oxygen uptake rate) time-domain variables. Models created to predict blood lactate concentration rate relied on variables reflecting interaction between front and back leg muscles, while parameters computed from front muscles and interactions between two legs were the most important variables for models created to predict oxygen uptake rate.© 2017 Elsevier Ltd.</p> +
Predicting the need for vehicle compressor repairs using maintenance records and logged vehicle data +
<p>Methods and results are presented for applying supervised machine learning techniques to the task of predicting the need for repairs of air compressors in commercial trucks and buses. Prediction models are derived from logged on-board data that are downloaded during workshop visits and have been collected over three years on large number of vehicles. A number of issues are identified with the data sources, many of which originate from the fact that the data sources were not designed for data mining. Nevertheless, exploiting this available data is very important for the automotive industry as means to quickly introduce predictive maintenance solutions. It is shown on a large data set from heavy duty trucks in normal operation how this can be done and generate a profit.</p><p>Random forest is used as the classifier algorithm, together with two methods for feature selection whose results are compared to a human expert. The machine learning based features outperform the human expert features, which supports the idea to use data mining to improve maintenance operations in this domain. © 2015 Elsevier Ltd.</p> +
<p>This paper presents an approach to designing an adaptive, data dependent, committee of multilayer perceptrons (MLP) for predicting trends (positive or negative change) of five financial attributes used for assessing future performance of a company. Total Asset Turnover (TAT), Current Ratio (CR), Gross Margin (GM), Operating Margin (OM), and Return on Equity (ROE) are the attributes used in this paper. A two- and three-years ahead prediction of change is considered. A Self-Organizing Map (SOM) used for data mapping and analysis enables building committees, which are specific (committee size and aggregation parameters) for each data point analyzed. When tested on data concerning 59 companies of the United States biotechnology sector, committees built according to the proposed technique outperformed both averaging and weighted averaging committees.</p> +
<p>This paper is concerned with a feature selection technique capable of generating an efficient feature set in a few selection steps. The feature saliency measure proposed is based on two factors, namely, the fuzzy derivative of the predictor output with respect to the feature and the similarity between the feature being considered and the feature set. The use of the fuzzy derivative enables modelling the vagueness that occurs in estimating the predictor output sensitivity. The feature similarity measure employed allows avoiding utilization of very redundant features. The experimental investigations performed on five real world problems have shown the effectiveness of the feature selection technique proposed. The technique developed removed a large number of features from the original data sets without reducing the classification accuracy of a classifier. In contrast, the accuracy of the classifiers utilizing the reduced feature sets was higher than those exploiting all the original features.</p> +
