Depression is a common mental disorder, which greatly affects the patients' daily life and work. Current depression detection relies almost exclusively on the clinical interview and structured questionnaire, consuming a lot of medical resources and risking a range of subjective biases. Our goal is to achieve a convenient and objective depression detection system, which can assist clinicians in their diagnosis of clinical depression. In this paper, we propose an experimental paradigm based on image cognition to record the reaction time data and eye movement data of the participants, build one of the largest datasets of depression. After extracting the
corresponding R-T (Reaction Time) features and E-M (Eye Movement) features that can reflect the participant’s attention bias, we use a standard classifier of Support Vector Machine to classify depressed patients and normal controls. Our method achieves accuracy up to 86%, which outperforms the previous related method. In our large-scale dataset, we get outstanding classification performance.