Analysis of Subjective Ratings of Expression Intensity for 3D Face Models Reconstructed from Monocular Images

We use facial expressions to communicate our emotions to others. In recent years, as 3D models have become more widely used in communication, it has become important to reflect more detailed facial expressions. In the field of computer vision, research has been conducted on 3D face models that reconstruct facial shapes, poses, and expressions from 2D images. For example, 3D face models have been developed such as DECA, which reconstructs individualized facial shapes, and EMOCA, which can reconstruct facial expressions in more detail. However, few studies have examined how humans perceive the emotional expressions of 3D models, and in particular, no studies have analyzed the intensity of facial expressions expressed by 3D models. Therefore, we analyzed subjective ratings of facial expression intensity for the 3D face models EMOCA and DECA reconstructed from monocular images. The experimental results showed that EMOCA represents the intensity of the facial expressions equivalent to that of 2D images for happiness, sadness, surprise, and anger. Although the experiment participants did not perceive any difference between EMOCA and DECA, EMOCA reproduced the intensity of facial expressions more accurately than DECA, and EMOCA had significantly higher facial expression intensity than DECA in happiness and disgust.