Output details

Texture and shape information fusion for facial expression and facial action unit recognition

<23> This paper proposed one of the first algorithms for facial expression recognition by fusing texture and shape information. Discriminant Non-negative Matrix Factorization (DNMF) was applied to the images, extracting the texture information, while Support Vector Machines (SVMs) were used to extract the shape information, in the form of grid deformations over time. Experiments were conducted using the Cohn Kanade database proving the superiority of using both sources of information (instead of using only texture or shape information) for the recognition of seven facial expressions and Facial Action Units (FAUs) recognition.