A Comparative Study of Facial Emotion Recognition with and without CLAHE in Varying Lighting Conditions

Facial emotion recognition (FER) plays a vital role in human-computer interaction, yet its accuracy can decrease under varying lighting conditions. This study conducts a comparative analysis of CNN-based FER models trained with and without the Contrast Limited Adaptive Histogram Equalization (CLAHE) preprocessing technique, using a combination of the FER2013 dataset and an additional Indonesian facial expression dataset. The experimental results show that applying CLAHE improves the weighted average F1score from 60 % to 64 %, outperforming not only the baseline CNN model but also previous studies, such as Gao (50%) and Saravanan (60 %), which used standard CNNs without image enhancement. These findings highlight the significant impact of preprocessing on model performance and demonstrate how comparative empirical evaluation can reveal the added value of simple enhancement methods like CLAHE in improving the accuracy and generalization of FER systems under real-world lighting variations.