Hybrid Separable Convolutional Inception Residual Network for Human Facial Expression Recognition

Facial expression recognition has been applied widely in human-machine interactions, security and business applications. The aim of facial expression recognition is to classify human expressions from their face images. In this work, we propose a novel neural network-based pipeline for facial expression recognition, Hybrid Separable Convolutional Inception Residual Network, using transfer learning with Inception residual network and depth-wise separable convolution. Specifically, our method uses multi-task convolutional neural network for face detection, then modifies the last two blocks of the original Inception residual network using depthwise separable convolution to reduce the computation cost, and finally utilizes transfer learning to take advantages of the transferable weights from a large face recognition dataset. Experimental result on three different databases - the Radboud Faces Database, Compounded Facial Expression of Emotions Database, and Real-word Affective Face Database, shows superior performance compared with the existing studies. Moreover, the proposed method is computationally efficient and reduces the trainable parameters by approximately 25% than the original Inception residual network.