Classifying range images of human faces with Hausdorff distance
B. Achermann, Horst Bunke · 2002
In this paper, an application of range image analysis in the area of face recognition is described. The process from image acquisition to recognition is presented in detail. Range images are captured with a multisensor system, merged in a fusion step, and preprocessed. Then, a canonical position of the sensed face is determined, and two different representation of the range data, based on point sets and voxel arrays, respectively, are computed. For the recognition step, a 3D version of the partial Hausdorff distance is introduced. Experiments on a database with 240 images of 24 persons yielded nearly perfect results.