An Artificial Intelligence Model for Sensing Affective Valence and Arousal from Facial Images

Area of Science:

• Psychology
• Computer Science
• Affective Computing

Background:

• Recent psychological research links facial expressions to dimensional affective states (valence and arousal).
• Existing AI models lack empirical data-driven approaches for estimating these affective states from facial imagery.

Purpose of the Study:

• To develop and validate a recurrent neural network-based AI model for estimating subjective valence and arousal from facial data.
• To provide the first distributable AI model for sensing affective valence and arousal based on an empirical database.

Main Methods:

• Developed a recurrent neural network (RNN) AI model.
• Trained the model using a database of participant affective states (valence/arousal) and corresponding facial images.
• Validated the model using leave-one-out cross-validation and a separate dataset of facial videos with affective ratings.

Main Results:

• The AI model demonstrated valid prediction of subjective valence and arousal states.
• Real-time prediction performance was comparable to commercial AI solutions like FaceReader.
• A graphical user interface was created for real-time affective state visualization from facial videos.

Conclusions:

• The developed AI model is the first empirically validated and distributable tool for sensing affective valence and arousal from facial data.
• Potential applications include mental health monitoring, marketing research, and human-computer interaction.