Affective Image Content Analysis: Two Decades Review and New Perspectives

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Visual data is rapidly growing, increasing the need for understanding image-induced emotions.
• Affective Image Content Analysis (AICA) aims to interpret the emotional impact of images.
• Key challenges include the affective gap, subjective perception, and data imperfections.

Purpose of the Study:

• To provide a comprehensive two-decade review of Affective Image Content Analysis (AICA).
• To highlight state-of-the-art methods addressing core AICA challenges.
• To identify future research directions in visual emotion AI.

Main Methods:

• Reviewing emotion representation models and datasets used in AICA.
• Summarizing and comparing handcrafted and deep features for emotion recognition.
• Analyzing learning methods for dominant, personalized, and distributed emotion prediction, including handling noisy data.

Main Results:

• Identified key emotion representation models and evaluated datasets for AICA.
• Compared various feature extraction techniques (handcrafted vs. deep).
• Detailed diverse learning strategies for AICA, including those robust to data noise and scarcity.

Conclusions:

• AICA has evolved significantly, with ongoing research addressing perception subjectivity and data quality.
• Future work should focus on deeper image understanding, group emotions, and viewer interaction.
• Advances in AICA promise more nuanced and personalized human-computer emotional experiences.