Facial expression recognition and its degree estimation

Machine Analysis of Facial Expressions

“…However, first steps toward this goal have been made. Some researchers described changes in facial expression that could be used to represent intensity variation automatically (Essa & Pentland, 1997;Kimura & Yachida, 1997;Lien et al, 1998), and an effort toward implicit encoding of intensity was reported by Zhang & Ji (2005). Automatic coding of intensity variation was explicitly compared to manual coding in Bartlett et al (2003a;.…”

Neuro Fuzzy Model for Human Face Expression Recognition

“…Detecting a unit set of action units for specific expression is not guaranteed. One promising approach for recognizing up to facial expressions intensities is to consider whole facial image as single pattern [4]. Kimura and his colleagues have reported a method to construct emotional space using 2D elastic net model and K-L expansions for real images [7].…”

Machine Learning Methods for Social Signal Processing

“…While FACS (Ekman et al 2002) provides a 5-point ordinal scale for coding the intensity AUs, there is no established standard for how to code the intensity of holistic facial expressions (e.g., those of the six basic emotions). Primarily for this reason and the observation in (Hess et al 1997) that the expression decoding accuracy and the perceived intensity of the underlying affective state vary linearly with the physical intensity of a facial display, the existing works on intensity estimation of facial expressions of the basic emotions resort to an unsupervised approach to modeling of the expression intensity (e.g., (Amin et al 2005, Shan 2007, Kimura & Yachida 1997, Lee & Xu 2003, Yang et al 2009b). The main idea in these works is that the variation in facial images due to the facial expressions can be represented on a manifold, where the image sequences are embedded as continuous curves.…”

“…Subsequently, Fuzzy K-Means was used to cluster the embeddings of each expression category into three fuzzy clusters corresponding to a low, moderate and high intensity of target expressions. (Kimura & Yachida 1997) used a Potential Net model to extract the motion-flow-based features from images of facial expressions, which were used to estimate a 2D eigenspace of the expression intensity. (Lee & Xu 2003) and (Yang et al 2009b) also performed the intensity estimation on a manifold of facial expressions.…”