Prototype-based Classifier for Automatic Diagnosis of Depressive Mood

摘要

Psychosocial disorders are a major public health issue, possibly leading to severe short and/or long-term consequences on both personal and professional levels. These troubles have to be diagnosed by a specialist doctor. However, Affective Computing (AC) can provide him with an assistance, as a both fast and inexpensive monitoring to the patient. Therefore, we propose a real-time automated tool for evaluating depressive moods, by way of a simple webcam observing the face. We have trained a classifier on AVEC2014 challenge database to extract prototypes of depressive faces, with respect to the Beck Depression Inventory II score (BDI-II). The system achieves with succes rates above 90%, an RMSE estimated at 4.30 and a realtime diagnosis abilitv.