Comparing the Effects of Various Preprocessing Techniques on the Performance of CNN for Facial Emotion Recognition

Facial Emotion Recognition has gained importance and relevance in all areas of human computer interaction. A large amount of research and scientific resources have been deployed for enhancing this domain. Despite this, it faces challenges arising from real-life imagery such as illumination variation, and expression intensity. The preprocessing stage is a performance determining factor that can be leveraged to overcome these challenges. This paper focuses on various preprocessing techniques and compares their performance on various available datasets for facial emotion recognition. The preprocessing techniques examined are Face Detection, Local Binary Patterns(LBP), Histogram Equalization, Edge Detection, Local Contrast Normalization (LCN), Global Contrast Normalization(GCN), and noise addition. The results obtained give adequate evidence to suggest the above techniques help improve the performance of CNN for the task of emotion classification.