Facial Emotion Analysis via Symmetrically Aligned Convolutional Neural Networks

Authors:Jacob Steinhardt,Stanford University

      Dario Amodei,Google Brain

Abstract: The analysis of facial expressions holds significant practical value across various domains, including healthcare, education, criminal investigation, transportation, and human-computer interaction. This study presents a novel model employing a Siamese Aligned Convolutional Neural Network (SACNN) for facial expression recognition. The SACNN model effectively segments facial images into left and right halves, subsequently performing expression recognition on these separate regions to achieve a notably high recognition accuracy. Evaluation on the FER2013 and CK+ datasets demonstrates an improvement in accuracy by 0.9% and 0.6% respectively. The experimental results affirm the efficacy of the proposed model in accurately identifying facial expressions.

Keywords:Expression Analysis; Siamese Neural Network; Feature Alignment; Segmentation