Enhancing Affective Modelling Through Preprocessing: A Comparative Review for Valence and Arousal Estimation

Abstract

Facial emotion recognition, especially the estimation of valence and arousal, plays a critical role in affective computing and human-computer interaction. The reliability of these predictions is significantly influenced by the preprocessing techniques used on facial images. This paper presents a comparative review of conventional and advanced preprocessing approaches, ranging from traditional image normalization to deep learning-driven methods, including data augmentation, multimodal fusion, and spatiotemporal encoding. Through critical analysis and referencing of recent literature, we highlight how preprocessing impacts model performance and suggest best practices for enhancing affective modeling pipelines.