Mouth open or closed decision for frontal face images with given eye locations

Determination of the open/closed state of a mouth is a desired feature considering the face ISO standard. In this paper a simple score is proposed for automatic mouth open/closed decision. Landmarks around the mouth are explored in order to calculate the score. Analysis of mouth location, scale and rotation introduced by nonreflective similarity transformations utilizing prior knowledge regarding the eye locations with the accompanied Jesorsky error is presented. Further, a novel system is proposed in which a search in the four dimensions; position, scale, rotation and shape variation is combined with a discriminative classifier in order to perform alignment of the landmarks. The system is evaluated on the XM2VTS and IMM face databases indicating the feasibility of the proposed system. The evaluation is presented utilizing a cumulative distribution of the mouth open/closed score error.