Person-Specific Joy Expression Synthesis with Geometric Method

The psychophysics of empathy: Using reverse-correlation to quantify the overlap between self & other representations of emotional expressions

“…While our participants' representations of affiliation smiles generally matched what was known from previous research (displacements in the mouth area corresponding to AU20-16-16-14 and in the eye areas corresponding to AU7-5-6-46; Figure 3-E), we found that each participant had a subtly idiosyncratic way to produce and represent smiling faces, involving various degrees of horizontal vs vertical lip motion (Figure 4-A), or eye widening vs squinting. While individual differences in the morphology of emotional expressions are a well-described phenomena in the computer science literature attempting to recognize or synthesize facial emotions [59,73], this phenomenon is relatively under-studied in the psychological literature [14]. Individual differences in smile morphology (e.g.…”

Combining GAN with reverse correlation to construct personalized facial expressions

“…On the other hand, local attributes are specific features or localized regions that can be individually modified to alter the facial appearance. These attributes focus on fine-grained details and localized features, such as hairstyle, presence of facial hair (beard, mustache), presence of accessories (glasses, earrings) [6,23,24], specific facial components (e.g., mouth shape [27][28][29], cheek, and eyebrows [29]) and facial muscles [3,18]. These attributes usually do not alter the entire facial structure and identity.…”

“…This approach achieved dynamically generating different local facial movements mostly around the mouth (e.g., bare teeth, high smile, lips up, mouth down, mouth extreme, mouth open). Zaied et al [27,28] proposed geometric and geometric-machine learning methods mainly to personalize smiles. However, these approaches [27][28][29] mainly manipulate the region around the mouth.…”

“…Zaied et al [27,28] proposed geometric and geometric-machine learning methods mainly to personalize smiles. However, these approaches [27][28][29] mainly manipulate the region around the mouth. Departing from these approaches, GANimation [3] can generate anatomically-aware face animation by taking a list of action units (AUs) as input.…”