A Fast and Robust Method for the Identification of Face Landmarks in Profile Images

摘要

Several applications in Computer Vision, like recognition, identification, automatic 3D modeling and animation and non conventional human computer interaction require the precise identification of landmark points in facial images. Here we present a fast and robust algorithm capable of identifying a specific set of landmarks on face profile images. First, the face is automatically segmented from the image. Then, the face landmarks are extracted. The algorithm is based on the local curvature of the profile contour and on the local analysis of the face features. The robustness of the presented approach is demonstrated by a set of experiments were ground truth data are compared with the result of our algorithm. A percentage of 92% correct identification and a mean error of 3.5 pixels demonstrate the robustness of the approach, which is of paramount importance for several applications.