Gaze-based Person Identification and Verification enable to recognize and to authenticate people by analyzing their gaze. In this thesis we carried out two kinds of experiments: in the first, we used, as stimuli, 30 symmetric images, while in the other we included 10 pattern graphics. The experiments were conducted on 37 testers and were repeated in three different sessions, separated each other by at least 24 hours. In the end, five testers for symmetric images and six testers for pattern graphics were excluded because of bad eye tracking data, thus resulting in a total of 32 testers and 31 testers, respectively. Features related to both single gaze samples acquired by the eye tracker and gaze fixations were employed. For all features, we computed different descriptive statistics. Also, we considered five different data sets, namely: 1) pattern graphics; 2) images characterized by a 360-degree symmetry; 3) images characterized by both vertical and horizontal symmetry; 4) images characterized by vertical symmetry; and 5) all images, characterized by all kinds of symmetry. Before processing, data underwent standardization and normalization processes, to improve the analysis results. Hyper-parameter tuning for the different classifiers employed was also performed. Support Vector Machine, Random Forest, Neural network, and Decision Tree were used as classifiers. Satisfactory identification and verification accuracies, sensitivities, specificities and good results for the area AUC under the ROC curves were obtained, which seem to suggest that the proposed approach is a feasible solution for soft biometrics (that does not require very high accuracies because the biometric method is used in addition to, and not instead of, other, more secure, authentication techniques). The idea behind this project is: that, for example, we could use gaze-based biometrics in ATM machines in addition to card and PIN, as a supplementary, soft biometric security level.
Biometria basata sul tracciamento oculare: uno studio sull'osservazione di immagini simmetriche e motivi regolari. L identificazione e la verifica basate sullo sguardo consentono di riconoscere e autenticare le persone analizzando il modo in cui guardano qualcosa. In questa tesi abbiamo condotto due tipi di esperimenti: nel primo abbiamo usato come stimoli 30 immagini simmetriche, mentre nell'altro abbiamo incluso 10 “pattern” (immagini con elementi grafici ripetuti). Gli esperimenti sono stati condotti su 37 tester e sono stati ripetuti in tre sessioni diverse, separate l'una dall'altra da almeno 24 ore. Alla fine, cinque tester per immagini simmetriche e sei tester per pattern sono stati esclusi a causa di dati di eye tracking scadenti, risultando in un totale di 32 tester e 31 tester, rispettivamente. Sono state utilizzate caratteristiche relative sia ai singoli campionamenti dello sguardo, sia alle fissazioni. Abbiamo calcolato diverse statistiche descrittive. Inoltre, abbiamo preso in considerazione cinque diversi set di dati, ottenutio: 1) immagini pattern; 2) immagini caratterizzate da una simmetria a 360 gradi; 3) immagini caratterizzate da simmetrie verticale e orizzontale; 4) immagini caratterizzate da simmetria verticale; e 5) tutte le immagini, caratterizzate da tutti i tipi di simmetria. Prima dell'elaborazione, i dati sono stati standardizzati e normalizzati, per migliorare i risultati dell'analisi. È stata eseguita anche la regolazione degli iperparametri per i diversi classificatori impiegati. I classificatori usati sono stati Support Vector Machines, Random Forest, Reti Neurali e Decision Tree. Identificazione e verifica hanno prodotto precisioni soddisfacenti, e anche buone sensibilita’ e specificita. Soddisfacenti anche i risultati per l'area AUC ( sottesa dalla curva ROC), il che sembra suggerire che l'approccio proposto è una soluzione fattibile per la soft biometrics (che non richiede precisioni molto elevate perché il metodo biometrico viene usato in aggiunta a, e non al posto di, altre tecniche di autenticazione). L'idea alla base di questo progetto è che, ad esempio, potremmo usare lo sguardo negli sportelli Bancomat, oltre a carta e PIN, come ulteriore livello di sicurezza.
Gaze-Based Biometrics: A Study on the Observation of Symmetric Images and Pattern Graphics
SIYUM, LETEBRHAN ALEMAYOH
2018/2019
Abstract
Gaze-based Person Identification and Verification enable to recognize and to authenticate people by analyzing their gaze. In this thesis we carried out two kinds of experiments: in the first, we used, as stimuli, 30 symmetric images, while in the other we included 10 pattern graphics. The experiments were conducted on 37 testers and were repeated in three different sessions, separated each other by at least 24 hours. In the end, five testers for symmetric images and six testers for pattern graphics were excluded because of bad eye tracking data, thus resulting in a total of 32 testers and 31 testers, respectively. Features related to both single gaze samples acquired by the eye tracker and gaze fixations were employed. For all features, we computed different descriptive statistics. Also, we considered five different data sets, namely: 1) pattern graphics; 2) images characterized by a 360-degree symmetry; 3) images characterized by both vertical and horizontal symmetry; 4) images characterized by vertical symmetry; and 5) all images, characterized by all kinds of symmetry. Before processing, data underwent standardization and normalization processes, to improve the analysis results. Hyper-parameter tuning for the different classifiers employed was also performed. Support Vector Machine, Random Forest, Neural network, and Decision Tree were used as classifiers. Satisfactory identification and verification accuracies, sensitivities, specificities and good results for the area AUC under the ROC curves were obtained, which seem to suggest that the proposed approach is a feasible solution for soft biometrics (that does not require very high accuracies because the biometric method is used in addition to, and not instead of, other, more secure, authentication techniques). The idea behind this project is: that, for example, we could use gaze-based biometrics in ATM machines in addition to card and PIN, as a supplementary, soft biometric security level.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/20.500.14239/24358