Estimation of Depression Anxieties and Stress through Clustering of Sequences of Visual and Thermal Face Images

Abstract

This paper presents an architecture for estimating the posterior probabilities of disorders in affective health by clustering sequences of emotional states. The disorders considered are depression, anxiety, and stress, based on the well-known DASS-21 model. The visual and thermal face image sequences elicit the emotional states through a proposed cascaded Convolutional Neural Network (CCNN) model. The framework of the proposed CCNN has 16 layers containing convolutional, pooling, and fully connected. The same architecture has been used to independently train the visual and thermal images to independently determine every seven emotional states. The sequence of the emotional states is clustered based on a pre-trained hidden Markov model (HMM). This model inputs a series of emotions obtained through face images and outputs the posterior estimates of stress, depression, and anxiety.