Understanding Image Memorability: Neural Correlates, Behavioral Characteristics, and Predictive Models

This paper presents a comprehensive review on the evolving research landscape of image memorability, aiming to synthesize existing knowledge and identify gaps in understanding the roles of emotional content, aesthetics, and visual salience. Through a systematic examination of recent studies and the critical analysis of the LaMem dataset, this review explores advancements in computational models for predicting image memorability, highlighting the shift towards classification-based approaches and the incorporation of semantic features. The findings reveal a complex interplay of factors influencing memorability and challenge the efficacy of traditional regression-based models, suggesting that the novel classification approach not only benchmarks but also, in some aspects, surpasses human consistency. By offering insights into the cognitive processes behind image memorability and proposing a novel computational model, this review contributes to a deeper understanding of the subject and outlines directions for future research, emphasizing the need for a holistic approach to studying image memorability.