Séminaire (organisé par l’équipe de recherche ICI)


Seconda Università di Napoli, Dipartimento di Psicologia and IIASS, Italy

A novel database for emotion recognition from handwriting

Résumé :

Recognizing emotions automatically is a recent domain due to the rapid development of human-machine interfaces. We present the first open handwriting database which relates emotions to handwriting. Affective data have been collected through a digitizing tablet.

The database includes samples of 48 subjects whose emotions, namely anxiety, depression and stress, are assessed by DASS scores. Seven tasks are recorded for each subject : pen- tagons and house drawing, words copied in handprint, circles and clock drawing, and sentence copied in cursive writing. Records consist in pen positions, both on-paper and in-air, pressure, time stamp and pen azimut. We report first experiments on this database. From an initial set of features extracted from both in-air and on-paper movements, relevant features for each targeted emotion are selected by a random forest approach. Emotions are then recognized from the relevant features. Our results, reported by OOB accuracy estimates, show that the 3 targeted emotions can be recognized with accuracies ranging from 59% to 69.8 %, anxiety being the emotion recognized with the highest accuracy.

Mardi 30 septembre 2014 à 14h00 en L200


FR SHIC 3272

Collegium UTC/CNRS