Landmark localization approach for facial computing

Face landmark localization is the technique of finding certain features on the face for subsequent face processing. With the recent growth in robust face recognition using facial components, facial landmark localization has proven to be an important phase for any subsequent processing on face images or facial components. Though it is conceptually very simple to locate facial landmarks like eyes, mouth, nose, it is challenging due to factors like illumination, occlusion, pose, inherent face variability and image conditions. Furthermore, other facial components like cheek bone area, forehead and nose-bridge provide more discriminant information between subjects. However, there is hardly any technique in the literature on how to detect these secondary regions. This paper proposes a technique that automatically detects a human face in an image and subsequently detect eyes, cheeks, nose, nose-bridge, mouth and forehead using Haar-cascade for detection of fiducial landmarks and spatial and geometric information of these fiducial landmarks for localization of secondary landmarks. The experiments are performed on FGNET-AD dataset and accuracy of above 90% was achieved for landmark localization.