Experiments on Global and Local Active Appearance Models for Analysis of Sign Language Facial Expressions

We explore features based on Active Appearance Modeling (AAM) of facial images within sign language videos. We employ a global AAM that initializes multiple local AAMs around places of interest. The local features offer a compact and descriptive representation of the facial regions of interest. The Global and Local AAM (GLAAM) is applied on Sign Language videos, and evaluated on classification experiments wrt. existing facial transcriptions of interest in data from the continuous sign language corpus of BU400 providing promising results.