Your search

In this paper, we focus on improving the age estimation accuracy on smartphones. Estimating a smartphone user’s age has several applications such as protecting our children online by filtering age-inappropriate contents, providing a customized e-commerce experience, etc. However, accuracy of the the state-of-the-art age estimation techniques that use touch behavior on smartphones is still limited because of the lack of sufficient amount of training data. We perform rigorous experiments using zoom gestures on smartphones and demonstrate that increasing the amount of training data can significantly improve the age estimation accuracy. Based on the findings in this study, we recommend creating a large touch dynamics-based age estimation data set so that more accurate age estimation models can be built and in turn, can be used more confidently.

A smartphone-based fall detection system has two major advantages over a traditional fall detection system that comes as a separate device: (1) the phone can automatically send messages to or call the emergency contact person when a fall is detected and (2) a user does not need to carry an extra device. This paper presents a novel two-step fall detection method which uses data extracted from smartphone sensors to detect falls. A fall can happen in many ways. A person can fall while he/she is walking, jogging, sitting, or even sleeping. Patterns of all falls are not the same. It is important to identify the type of falls to precisely distinguish it from non-falls (normal activities). Hence, our method first identifies the correct type of falls by performing multi-class classification. In the second step, this method produces a binary decision based on the multiclass prediction. We collected data from 10 users to evaluate our proposed fall detection method. Each user performed five normal activities-namely, walking, jogging, standing, sitting, lying, and also fell after performing each activity. We performed experiments with five common smartphone sensors: accelerometer, gyroscope, magnetometer, gravity, and linear acceleration. We tested five machine learning classifiers-namely, Support Vector Machine, K-Nearest Neighbor, Decision Tree, Random Forest, and Naive Bayes. Our two-step fall detection method achieved the maximum accuracy of 95.65% and the maximum area under ROC curve (AUC) of 0.93, both with the gyroscope sensor and Support Vector Machine classifier.

Image creation and retention are growing at an exponential rate. Individuals produce more images today than ever in history and often these images contain family. In this paper, we develop a framework to detect or identify family in a face image dataset. The ability to identify family in a dataset of images could have a critical impact on finding lost and vulnerable children, identifying terror suspects, social media interactions, and other practical applications. We evaluated our framework by performing experiments on two facial image datasets, the Y-Face and KinFaceW, comprising 37 and 920 images, respectively. We tested two feature extraction techniques, namely PCA and HOG, and three machine learning algorithms, namely K-Means, agglomerative hierarchical clustering, and K nearest neighbors. We achieved promising results with a maximum detection rate of 94.59% using K-Means, 89.18% with agglomerative clustering, and 77.42% using K-nearest neighbors. © 2020 World Scientific Publishing Company.

In this paper, we develop an indoor positioning system using smartphones. An indoor positioning system plays a vital role in indoor spaces such as home, office, university, airport, and hospital buildings by locating and tracking persons, devices, and assets. Our indoor positioning system is applicable in any indoor spaces which has smart devices such as smartphones, tablets, smartwatches, and robots with a Wi-Fi connection. We used Wi-Fi-based fingerprinting technique t o build o ur indoor positioning system because a Wi-Fi-based system can leverage existing Wi-Fi infrastructure and hence, it is cost effective. A major challenge in implementing a Wi-Fi-based fingerprinting technique is the missed access points (APs) problem. In this paper, we address this critical challenge by proposing a localization procedure called ‘cosine similarity + k-means clustering'. In this localization procedure, we leverage k-means clustering algorithm in identifying the wrong location estimates produced by the cosine similarity measure because of missed APs problem. To evaluate the effectiveness of our proposed localization procedure, we collected data from three different scenarios, specifically, home, office, a nd university f or creating signal m ap a nd performing localization tests. Additionally, we tested both stationary and walk data. Our experimental results prove that our ‘cosine similarity + k-means clustering’ localization procedure is effective in mitigating the detrimental impact of missed APs, and consequently, it significantly improves localization accuracy.

Design of a serial fusion based multi-biometric verification system requires fixing several parameters, such as reject thresholds at each stage of the architecture and the order in which each individual verifier is placed within the multi-stage system. Selecting the order of verifier is a crucial parameter to fix because of its high impact on verification errors. A wrong choice of verifier order might lead to tremendous user inconvenience by denying a large number of genuine users and might cause severe security breach by accepting impostors frequently. Unfortunately, this design issue has been poorly investigated in multi-biometric literature. In this paper, we address this design issue by performing experiments using three different serial fusion based multi-biometric verification schemes. We did our experiments on publicly available NIST multi-modal dataset. We tested 24 orders—all possible orders originated from four individual verifiers—on a four-stage biometric verification system. Our experimental results show that the verifier order “best-to-worst”, where the best performing individual verifier is placed in the first stage, the next best performing individual verifier is placed in the second stage, and so on, is the top performing order. In addition, we have proposed a modification to the traditional architecture of serial fusion based multi-biometric verification systems. With rigorous experiments on the NIST multi-modal dataset and using three serial fusion based multi-biometric verification schemes, we demonstrated that our proposed architecture significantly improves the performance of serial fusion based multi-biometric verification systems.

As more people rely on smartphones to store sensitive information, the need for robust security measures is all the more pressing. Because traditional one shot authentication methods like PINs and passwords are vulnerable to various attacks, we present a behavioral biometrics based smartphone authentication system using swipes. While previous research focused on a single kind of swipe, our data set features swipes using different fingers and directions collected from 36 users across three sessions. In our system, we experimented with support vector machine (SVM) and random forest (RF) classifiers. We investigated which finger, direction, and classifier provided the best individual swipe authentication results. Then, we analyzed whether fusion of different fingers and directions improved results. The best unimodal result came from a rightward swipe with right thumb using SVM, which resulted in an area under ROC curve (AUC) of 0.936 and an equal error rate (EER) of 0.135. We found that swipes using thumbs offered better performance. Fusion improves results for the most part, and our best result was the combination of a leftward swipe with right thumb and a leftward swipe with left thumb. This combination gave an AUC of 0.969 and EER of 0.081 with the SVM classifier.

We investigate the viability of the capacitive swipe gesture as a biometric modality. While the regular swipe gesture and the capacitive image have been widely explored in biometric literature, the capacitive swipe gesture is fairly new in this line of research. To our knowledge, only one recent study has explored the capacitive swipe gesture, and demonstrated its promise. However, that study is limited by a number of factors, such as using a very small data set in the experiments, collecting data in a single session, allowing the same impostor in both training and testing phases of authentication models, etc. In our paper, we address all these limitations, and rigorously explore the capacitive swipe gesture by creating a new large data set. Additionally, we develop a new technique to preprocess capacitive swipe gesture data, and demonstrate its effectiveness by comparing with existing techniques. A large set of experiments with four machine learning classifiers and two swipe directions prove that the capacitive swipe gesture can be effectively used for user authentication in smartphones.

In this paper, we develop a new point-of-entry security measure for smartphone users. We devise a concept, the “Quad Swipe Pattern”, which includes four swipes from a user in four directions and utilizes the user’s swipe behavior for authentication. The Quad Swipe Pattern overcomes several shortcomings present in current point-of-entry security measures. We performed several experiments to demonstrate the effectiveness of the Quad Swipe Pattern in smartphone user authentication. We evaluated the Quad Swipe Pattern using five machine learning classifiers, three datasets of different sizes, and five different fingers. In addition, we studied how fusion of information from multiple fingers and multiple classifiers can improve the performance of Quad Swipe Pattern. All of our experimental results show significant promise of the Quad Swipe Pattern as a new point-of-entry security measure for smartphones. With a Neural Network model, the Quad Swipe Pattern achieves the Accuracy of 99.7%, False Acceptance Rate of 0.4%, and False Rejection Rate of 0%. With Support Vector Machine, the Quad Swipe Pattern achieves the Accuracy of 99.5%, False Acceptance Rate of 0.4%, and False Rejection Rate of 1.7%. With fusion of two best fingers, the Quad Swipe Pattern demonstrates an excellent performance of a zero Equal Error Rate.

We incorporate deep learning techniques into capacitive images of body parts (ear, four fingers, and thumb) to improve the performance of user authentication in smartphones. Use of a capacitive touchscreen as an image sensor has several advantages, such as it is less sensitive to poor illumination conditions, occlusions, and pose variations. Also, it does not need an additional hardware like iris or fingerprint scanner. Use of capacitive images for user authentication is not new. However, the performance, specially, false reject rates (FRRs) of the state-of-the-art capacitive image-based systems are poor. In this paper, we focus on improving the performance and leverage deep learning. Deep learning techniques demonstrated spectacular performance in previous physical biometrics-based research. However, to our knowledge, effectiveness of deep learning is still unexplored in capacitive touchscreen-based user authentication. In order to bridge this research gap, we devise a multi-modal deep learning model, namely UASNet, and compare its performance with a large set of uni- and multi-modal baselines. Using the UASNet, we achieve an accuracy of 99.77%, an EER of 0.48%, and an FRR of 1.19% at FAR of 0.06%.

We outline a novel method of user authentication for smart mobile devices, such as smartphones or tablets and propose movement pattern based authentication as an alternate to current methods that relies on a pin or drawn-pattern. While the current methods are vulnerable against common attacks (e.g., smudge attacks, shoulder surfing), our method, in contrast, is more resilient against the attacks of these kinds because it utilizes sensory data given off by the device during a preset movement for authentication. In our experiment, we recorded the values given off by four physical observational sensors: (1) accelerometer, (2) linear accelerometer, (3) gyroscope and (4) tilt sensor, which each had three axes, over a set of movements. We experimented with 10 arbitrary movement-patterns and gathered 12 samples of each (net 120 samples) to test with. We developed our own method of authentication, through which we performed 35,650 authentication attempts and found a 20.36% Equal Error Rate.

A multistage biometric verification system uses multiple biometrics and/or multiple biometric verifiers to generate a verification decision. The core of a multistage biometric verification system is reject option which allows a stage not to give a genuine/impostor decision when it is not confident enough. This paper studies the effectiveness of symmetric rejection for multistage biometric verification systems. The symmetric rejection method determines the reject region by symmetrically rejecting equal proportion of genuine and impostor scores. The applicability of a multistage biometric verification system depends on how secure and user convenient it is, which is measured by the performance–cost trade-off. This paper analyzes the performance–cost trade-off of symmetric rejection method by conducting extensive experiments. Experiments are performed on two biometric databases: (1) publicly available NIST database and (2) a keystroke database. In addition, the symmetric rejection method is empirically compared with two existing rejection methods: (1) sequential probability ratio test-based method, which uses score-fusion and (2) Marcialis et al.’s method, which does not use score fusion. Results demonstrate strong effect of symmetric rejection method on creating a secure and user convenient multistage biometric verification system.

In this paper, we study the influence from the sentiment of regular tweets on retweeting. We propose a method to calculate the sentiment score for each tweet and each Twitter user. This method enables us to place the tweets and retweets into the same time period to explore the sentiment factor. We adopt the correlation coefficient between the sentiment scores of regular tweets and those of retweets to measure the influence. We categorize the Twitter users in three different ways to investigate three factors, which are the number of followers, betweenness centrality and the types of accounts. Community detection and machine learning are integrated into our approach. We find that the difference for correlation coefficients exists between different levels of the number of followers, and different types of users. Our method sheds a light on better predicting the dynamics of tweets diffusion by including the sentiment factor into the prediction model. © 2017 IEEE.

In online social networks (OSN), followers count is a sign of the social influence of an account. Some users expect to increase the followers count by following more accounts. However, in reality more followings do not generate more followers. In this paper, we propose a two player follow-unfollow game model and then introduce a factor for promoting cooperation. Based on the two player follow-unfollow game, we create an evolutionary follow-unfollow game with more players to simulate a miniature social network. We design an algorithm and conduct the simulation. From the simulation, we find that our algorithm for the evolutionary follow-unfollow game is able to converge and produce a stable network. Results obtained with different values of the cooperation promotion factor show that the promotion factor increases the total connections in the network especially through increasing the number of the follow follow connections.

In this paper, we provide a consistent, inexpensive, and easy to use graphical user interface (GUI) smart phone application named Sleep Apnea Screener (SAS) that can diagnosis Obstructive Sleep Apnea (OSA) based on demographic data such as: gender, age, height, BMI, neck circumference, waist, etc., allowing a tentative diagnosis of OSA without the need for overnight tests. The developed smart phone application can diagnosis sleep apnea using a model trained with 620 samples collected from a sleep center in Corpus Christi, TX. Two machine learning classifiers (i.e., Logistic Regression (LR) and Support Vector Machine (SVM)) were used to diagnosis OSA. Our preliminary results show that at-home OSA screening is indeed possible, and that our application is effective method for covering large numbers of undiagnosed cases.

Selecting the order of verifier in a serial fusion based multi-biometric system is a crucial parameter to fix because of its high impact on verification errors. A wrong choice of verifier order might lead to tremendous user inconvenience by denying a large number of genuine users and might cause severe security breach by accepting impostors frequently. Unfortunately, this design issue has been poorly investigated in multi-biometric literature. In this paper, we address this design issue by performing experiments using three different serial fusion based multi-biometric verification schemes, in particular (1) symmetric scheme, (2) SPRT-based scheme, and (3) Marcialis et al.’s scheme. We experimented on publicly available NIST-BSSR1 multi-modal database. We tested 24 orders—all possible orders originated from four individual verifiers—on a four-stage biometric verification system. Our experimental results show that the verifier order “best-to-worst”, where the best performing individual verifier is placed in the first stage, the next best performing individual verifier is placed in the second stage, and so on, is the top performing order for all three serial fusion schemes mentioned above.

Twitter users often crave more followers to increase their social popularity. While a variety of factors have been shown to attract the followers, very little work has been done to analyze the mechanism how Twitter users follow or unfollow each other. In this paper, we apply game theory to modeling the follow-unfollow mechanism on Twitter. We first present a two-player game which is based on the Prisoner’s Dilemma, and subsequently evaluate the payoffs when the two players adopt different strategies. To allow two players to play multiple rounds of the game, we propose a multi-stage game model. We design a Twitter bot analyzer which follows or unfollows other Twitter users by adopting the strategies from the multi-stage game. We develop an algorithm which enables the Twitter bot analyzer to automatically collect and analyze the data. The results from analyzing the data collected in our experiment show that the follow-back ratios for both of the Twitter bots are very low, which are 0.76%0.76%0.76\% and 0.86%0.86%0.86\%. This means that most of the Twitter users do not cooperate and only want to be followed instead of following others. Our results also exhibit the effect of different strategies on the follow-back followers and on the non-following followers as well.

Obstructive sleep apnea (OSA) is a well-known sleep ailment. OSA mostly occurs due to the shortage of oxygen for the human body, which causes several symptoms (i.e., low concentration, daytime sleepiness, and irritability). Discovering the existence of OSA at an early stage can save lives and reduce the cost of treatment. The computer-aided diagnosis (CAD) system can quickly detect OSA by examining the electrocardiogram (ECG) signals. Over-serving ECG using a visual procedure is challenging for physicians, time-consuming, expensive, and subjective. In general, automated detection of the ECG signal’s arrhythmia is a complex task due to the complexity of the data quantity and clinical content. Moreover, ECG signals are usually affected by noise (i.e., patient movement and disturbances generated by electric devices or infrastructure), which reduces the quality of the collected data. Machine learning (ML) and Deep Learning (DL) gain a higher interest in health care systems due to its ability of achieving an excellent performance compared to traditional classifiers. We propose a CAD system to diagnose apnea events based on ECG in an automated way in this work. The proposed system follows the following steps: (1) remove noise from the ECG signal using a Notch filter. (2) extract nine features from the ECG signal (3) use thirteen ML and four types of DL models for the diagnosis of sleep apnea. The experimental results show that our proposed approach offers a good performance of DL classifiers to detect OSA. The proposed model achieves an accuracy of 86.25% in the validation stage.

Weaknesses in smartphone security pose a severe privacy threat to users. Currently, smartphones are secured through methods such as passwords, fingerprint scanners, and facial recognition cameras. To explore new methods and strengthen smartphone security, we developed a capacitive swipe based user authentication and identification technique. Swipe is a gesture that a user performs throughout the usage of a smartphone. Our methodology focuses on using the capacitive touchscreen to capture the user's swipe. While the user swipes, a series of capacitive frames are captured for each swipe. We developed an algorithm to process this series of capacitive frames pertaining to the swipe. While different swipes may contain different numbers of capacitive frames, our algorithm normalizes the frames by constructing the same number of frames for every swipe. After applying the algorithm, we transform the normalized frames into gray scale images. We apply principal component analysis (PCA) to these images to extract principal components, which are then used as features to authenticate/identify the user. We tested random forest (RF) and support vector machine (SVM) algorithms as classifiers. For authentication, the performance of SVM (tested with left swipes) was more promising than RF, yielding a maximum accuracy of 79.88% with an FAR and FRR of 15.84% and 50%, respectively. SVM (tested with right swipes) produced our maximum identification accuracy at 57.81% along with an FAR and FRR of 0.60% and 42.18%, respectively. © 2020 IEEE.