Handwriting based Personality Identification using Textural Features

Nowadays, personality identification based on handwriting processing is becoming a very active field of research and experimentation, for the reason of its high demand in domains such as resources management, criminal investigations and mental health diagnostics. The implicit information includes attributes like writer identity, gender, age group and handedness etc. Furthermore, handwriting has also been employed as indicator of neurodegenerative disorders, cognitive and emotional development and the personality attributes of writers. Several works have revealed the existence of a significant relationship between people's psychological traits and their manuscript writings. This paper aim to present a technique that exploits textural features extracted from handwriting images for the purpose of predicting the writer's personality traits. Mainly, the work is based on the HWxPI Dataset which was built as a part of the multimedia information processing for personality & social networks analysis (MIPPSNA) contest. The Five Factor Model (FFM) was employed to assign binary labels to individuals on five personality traits and the writing samples of the same individuals were employed for prediction of the same using automated techniques. We adopt a global approach and represent the handwriting images using the Oriented Basic Image Features Column (oBIFC) scheme. Being a multi-label classification problem, we employ RankSVM to predict the labels (0 or 1) for each of FFM traits. Experimental study of the system was carried out using the protocol of the MIPPSNA contest as well as using a k-fold cross validations. The realized results validated the effectiveness of the proposed technique in characterizing personality traits from the visual information in handwriting.