Detection of facial parts via deformable part model using part annotation

In this paper, we propose a novel method for facial parts detection based on Deformable Part Model (DPM). In DPM, the parts are useful regions to detect the face and do not always correspond to the facial parts such as eye, nose and mouth. We model facial parts as a part filter and use annotation to training the position and size. In addition, we discuss the algorithm to deal with the variation of bounding box in the annotation. Our experimental results show that the proposed algorithm improves DPM for facial parts detection.