Facial Landmark-based Approach for Classification of Children’s Facial Expressions and Facial Action Codes
Description/Abstract/Artist Statement
Classification of facial expressions from children is a relatively unexplored area in the literature as few facial expression datasets taken of children exist. To explore and understand the classification of children’s facial expressions, images from the Child Affective Face Set (CAFE) are classified by facial expression using features based on facial landmark points. One limitation of CAFE is that it is not encoded with the Facial Action Coding System (FACS), which describes facial movements based on anatomical analysis. To address this limitation, we also propose a method for FACS classification of the CAFE dataset. The Discriminative Response Map Fitting (DRMF) toolbox is used to extract 66 landmark points on the face. Interpoint distance features are computed as the Euclidean distance between facial landmark point pairs. Interpoint distance features are then scaled between [-1.0, 1.0]. Multi-class Support Vector Machine (SVM) is used to classify two subsets of images from the CAFE dataset: the full dataset and a reduced dataset removing images where the child posed with mouth open or tongue protrusion. Grid search is used to identify appropriate parameters for the SVM classifier. The FACS-encoded the Extended Cohn-Kanade Facial Expression Set (CK+) is used to train a FACS classier using a combination of interpoint distance and texture features.
Faculty Advisor/Mentor
Dr. Khan Iftekharuddin
Presentation Type
Poster
Disciplines
Other Electrical and Computer Engineering
Session Title
Poster Session
Location
Learning Commons @ Perry Library, Northwest Atrium
Start Date
2-3-2018 8:00 AM
End Date
2-3-2018 12:30 PM
Facial Landmark-based Approach for Classification of Children’s Facial Expressions and Facial Action Codes
Learning Commons @ Perry Library, Northwest Atrium
Classification of facial expressions from children is a relatively unexplored area in the literature as few facial expression datasets taken of children exist. To explore and understand the classification of children’s facial expressions, images from the Child Affective Face Set (CAFE) are classified by facial expression using features based on facial landmark points. One limitation of CAFE is that it is not encoded with the Facial Action Coding System (FACS), which describes facial movements based on anatomical analysis. To address this limitation, we also propose a method for FACS classification of the CAFE dataset. The Discriminative Response Map Fitting (DRMF) toolbox is used to extract 66 landmark points on the face. Interpoint distance features are computed as the Euclidean distance between facial landmark point pairs. Interpoint distance features are then scaled between [-1.0, 1.0]. Multi-class Support Vector Machine (SVM) is used to classify two subsets of images from the CAFE dataset: the full dataset and a reduced dataset removing images where the child posed with mouth open or tongue protrusion. Grid search is used to identify appropriate parameters for the SVM classifier. The FACS-encoded the Extended Cohn-Kanade Facial Expression Set (CK+) is used to train a FACS classier using a combination of interpoint distance and texture features.