Facial Expression Recognition Using Convolutional Neural Network Through Region-based Patch Generation: Harnessing Subtle Facial Cues

The task of classifying human emotions based on facial expressions has been a challenging area of research for the past two decades. Early approaches relied on hand-crafted features such as SIFT, HOG, and LBP and classifiers trained on facial expression datasets. However, these methods failed to provide accurate results on newer, more complex datasets. Recently, the trend has shifted towards using deep learning models for facial expression classification, providing improved results. Nevertheless, there is still room for improvement in the accuracy of these models. In this research, we propose a novel model to improve facial expression classification accuracy on several benchmark datasets. Our model is based on generating nine different patches for different relevant facial regions and uses nine different streams, each for one of those micro-facial regions. We evaluated our model on the CK+, Oulu-Casia, and RAF-DB datasets and achieved state-of-the-art results, including 10-fold cross-validation accuracy of 99.8% and 91.25% for CK+ and Oulu-Casia datasets, respectively. Our proposed model demonstrates a significant improvement in facial expression classification accuracy and provides a promising approach to future research in this field.