Enhancing facial expression recognition in uncontrolled environment: a lightweight CNN approach with pre-processing

Facial expressions play a key role in human non-verbal type of communication, providing key insights into emotions and intentions. These expressions serve as universal signals, helping individuals convey their internal states across various personal and social contexts. With the growing interest in automatic facial emotion recognition, deep neural networks have emerged as a popular approach for detecting human emotions, even under challenging, real-world conditions. However, external factors can affect the system's performance, degrading the quality of facial features and making emotion detection more difficult. In the presented paper, we propose a highly optimized lightweight convolutional neural network (LCNN) for emotion recognition in controlled and uncontrolled environments. The proposed model is designed to learn hidden nonlinear patterns from facial images. The proposed convolutional neural network consisting a series of convolutional layers followed by max-pooling layers. The model's performance is evaluated with and without pre-processing steps to highlight the importance of pre-processing in improving detection accuracy. The LCNN achieves 65% accuracy on the FER-2013 dataset and 98% on the CK + dataset.