Local Second-Order Gradient Cross Pattern for Automatic Depression Detection

Depression is a psychiatric disorder that seriously affects people's work and life. At present, the development of the automatic depression detection technology has become the focus of many researchers due to the serious imbalance of doctor-patient ratio. Physiological studies have revealed that there are differences in facial activity between normal and depressed individuals, so some works has been done to detect depression by extracting facial features. However, these works are limited in capturing the subtle changes. For these reasons, this paper proposes a novel local pattern named Local Second-Order Gradient Cross Pattern (LSOGCP) to extract the subtle facial dynamics in videos to improve the accuracy of depression detection. In particular, we firstly obtain LSOGCP feature through high-order gradient and cross coding scheme to characterize the detailed texture of each frame. Then LSOGCP histograms from three orthogonal planes (TOP) are generated to form the video representation denoted as LSOGCP-TOP. Finally, a hierarchical method of between-group classification and within-group regression is employed to predict the score of depression severity. Experiments are conducted on two publicly available databases i.e. AVEC2013 and AVEC2014. The results demonstrate that our proposed method achieves better performance than the previous algorithms.