The work presented in this thesis consists of a study on the potentials of eye data for the implementation of a biometric systems, namely to establish the identity of a subject relying only on his or her physical or behavioral attributes. We have exploited both features related to strictly physical aspects of the eye (i.e., pupil size), and some others which depend more specifically on the ocular behavior of the individual (i.e., gaze coordinates, fixations and saccades). In our work, we have considered the two biometric processes traditionally employed to check the user’s identity: identification, i.e. determining the identity of an individual, and verification, i.e. validating an a-priori identity claim. The experiment was designed using the OGAMA software, and consisted in a presentation of 18 images belonging to three different “affective” categories (six ‘positive’ images, six ‘negative’ images and six ‘neutral’ images). Tests involved a total of 30 subjects, and were repeated in three different sessions. The images were displayed on a screen, each for six seconds, in a random sequence, and participants were not instructed with any specific task to accomplish, but they just had to look at them freely. An Eye Tribe eye tracker, with a sampling rate of 30 Hz, was employed for the acquisition of the eye data during the observation, and, again, the OGAMA software was used to record and analyze gaze data. Regarding the analysis, we followed two main approaches: the first, focusing on the image as a whole and considering all the data; the second, taking into account only data related to some Areas Of Interests, specially defined. These two methods were applied to both identification and verification, initially considering all the 18 images, and then also analysing six images at time (each single category). The Orange toolkit was used for the classification process, along with various machine learning algorithms (namely, Logistic Regression, AdaBoost, Random Forest and Neural Network). Overall, we have reached encouraging results that suggest that eye data can be used for gaze-based biometrics. This approach is considered a sort of ”soft biometrics”, i.e. it does not require extremely high identification or verification rates.
Biometria basata sul tracciamento oculare: uno studio sull’osservazione libera di immagini con contenuto "emozionale". Il lavoro presentato in questa tesi consiste in uno studio sulle potenzialità offerte dai dati oculari per l’implementazione di un sistema biometrico, vale a dire per stabilire l’identità di un soggetto basandosi solamente sui suoi attributi fisici o comportamentali. Abbiamo sfruttato alcune caratteristiche legate agli aspetti strettamente fisici dell'occhio (in particolare, la dimensione della pupilla) e altre che dipendono in modo più specifico dal comportamento oculare dell'individuo (cioè, le coordinate dello sguardo, le fissazioni e le saccadi). Nel nostro lavoro, abbiamo preso in considerazione i due processi biometrici tradizionalmente impiegati per verificare l'identità di un utente: l’identificazione, cioè il riconoscimento di un individuo all’interno di un insieme e la verifica, ossia la convalida di una dichiarazione di identità. L'esperimento è stato progettato utilizzando il software OGAMA e consiste in una presentazione di 18 immagini appartenenti a tre diverse categorie (6 immagini “positive”, 6 immagini “negative” e 6 immagini “neutre”). Ai test, suddivisi in tre sessioni, hanno partecipato 40 soggetti. Le immagini erano mostrate su uno schermo, ciascuna per sei secondi, in una sequenza casuale, e ai partecipanti non era stato affidato alcun compito specifico da svolgere, ma erano semplicemente stati invitati ad osservare liberamente le immagini. Un eye tracker Eye Tribe, con una frequenza di campionamento di 30 Hz, è stato impiegato per l’acquisizione dei dati oculari, e il software OGAMA è stato nuovamente utilizzato per registrare ed analizzare i dati. Per quanto riguarda l’analisi, abbiamo seguito due approcci principali: il primo, concentrandoci sull’immagine nel suo complesso e tenendo in considerazione tutti i dati; il secondo, considerando solamente i dati relativi a delle Aree di Interesse appositamente definite. Questi due metodi sono stati applicati sia all’identificazione sia alla verifica, prima considerando tutte le 18 immagini e successivamente anche analizzando 6 immagini per volta (ogni singola categoria). Per il processo di classificazione è stato impiegato il software Orange, insieme a diversi algoritmi di machine learning (in particolare, Logistic Regression, AdaBoost, Random Forest e Neural Network). Nel complesso, abbiamo raggiunto risultati incoraggianti, che suggeriscono che i dati oculari possano essere utilizzati per la biometria basata sul tracciamento oculare. Questo approccio è considerato un tipo di “biometria soft”, che quindi non richiede il raggiungimento di tassi di riconoscimento estremamente elevati.
Gaze-Based Biometrics: A Study on Free Observation of Images with “Affective” Content.
BETTINI, ELISA
2016/2017
Abstract
The work presented in this thesis consists of a study on the potentials of eye data for the implementation of a biometric systems, namely to establish the identity of a subject relying only on his or her physical or behavioral attributes. We have exploited both features related to strictly physical aspects of the eye (i.e., pupil size), and some others which depend more specifically on the ocular behavior of the individual (i.e., gaze coordinates, fixations and saccades). In our work, we have considered the two biometric processes traditionally employed to check the user’s identity: identification, i.e. determining the identity of an individual, and verification, i.e. validating an a-priori identity claim. The experiment was designed using the OGAMA software, and consisted in a presentation of 18 images belonging to three different “affective” categories (six ‘positive’ images, six ‘negative’ images and six ‘neutral’ images). Tests involved a total of 30 subjects, and were repeated in three different sessions. The images were displayed on a screen, each for six seconds, in a random sequence, and participants were not instructed with any specific task to accomplish, but they just had to look at them freely. An Eye Tribe eye tracker, with a sampling rate of 30 Hz, was employed for the acquisition of the eye data during the observation, and, again, the OGAMA software was used to record and analyze gaze data. Regarding the analysis, we followed two main approaches: the first, focusing on the image as a whole and considering all the data; the second, taking into account only data related to some Areas Of Interests, specially defined. These two methods were applied to both identification and verification, initially considering all the 18 images, and then also analysing six images at time (each single category). The Orange toolkit was used for the classification process, along with various machine learning algorithms (namely, Logistic Regression, AdaBoost, Random Forest and Neural Network). Overall, we have reached encouraging results that suggest that eye data can be used for gaze-based biometrics. This approach is considered a sort of ”soft biometrics”, i.e. it does not require extremely high identification or verification rates.È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/24256