Emotional State Prediction from Online Handwriting and Signiture Biometrics /
Yasemin BAY; Supervisor: Assoc. Prof. Dr. Erbuğ ÇELEBİ
- 115 sheets; 30 cm.
Thesis (Ph.D) - Cyprus International University. Institute of Graduate Studies and Research Management Information Systems Department
Includes bibliography (sheets 95-111)
ABSTRACT It has long been known that handwriting reveals many information for an individual's identification. For the last century, investigators first tried to predict the owner of the handwriting then recognized that it reveals more information such as the characteristics of the owner therefore shifted most of their attention to personal characteristics estimation, such as emotional state, However, to be able to evaluate and investigate the emotional state of an individual there is a need of a database with correctly labelled data with the relevant ground truth information. Yet, for researchers to be able to evaluate one's emotional state fairly is to work on a database providing the basic requirements which are, including contributors' demographic labels (age, gender, ethnicity etc.) or emotional status in addition to their identification labels. Unfortunately, there are almost no database, neither public nor commercial, providing this useful information. This dissertation will first propose a novel and rigorous handwriting and signature biometric database both in offline and online form and then predict one’s emotional state. To achieve this goal wide range of ground truths (i.e. identity labels, demographic labels as well as emotional state labels) will be used which are not contemplated by any other publicly or commercially available biometric handwriting and signature databases in the literature. Therefore, making this novel database available will shed a light to the biometric research community once again and researchers can actively work on it to contribute to science. The proposed database comprised total of 134 volunteers with handwriting and signature biometric data acquired using a digitizing tablet and a stylus pen. The process consisted of four different steps, each with aiming to extract different emotional states Task 1 - natural, Task 2 - happy, Task 3 - sad and Task 4 - stress. The database presented also includes individuals’ demographic information such as age, gender, handedness, education level and nationality. Database in total consists of 804 handwriting and 8040 signature biometric samples. To be able to achieve the goal of this dissertation there have been 3 different experiments carried out to predict emotional state from handwriting and signature biometrics and the iii results obtained were encouraging with slightly high recognition rates and proves that both handwriting and signature biometrics contain valuable information regarding one’s emotional state. Out of 42 features 11 of them were selected after the feature selection method for classification purpose.