No abstract

This paper presents two novel facial expression recognition techniques: the real-time ensemble for facial expression recognition (REFER) and the facial expression recognition network (FERNet). Both approaches can detect facial expressions from various poses, distances, angles, and resolutions, and both techniques exhibit high computational efficiency and portability. REFER outperforms the existing approaches in terms of cross-dataset accuracy, making it an ideal network to use on fresh data. FERNet is a compact convolutional neural network that uses both geometric and texture features to achieve up to 98% accuracy on the MUG dataset. Both approaches can process 14 frames per second (FPS) from a live video capture on a battery-powered Raspberry Pi 4.INDEX TERMS Facial expression recognition, machine learning, multithreaded, active shape model, poseinvariant.

No abstract

No abstract