Nasal Region-Based 3D Face Recognition under Pose and Expression Variations Hamdi Dibeklio˘ glu 1 , Berk G¨ okberk 2 , and Lale Akarun 1 1 Bo˘ gazi¸ci University, Computer Engineering Department, Turkey {hamdi.dibeklioglu,akarun}@boun.edu.tr 2 University of Twente, Department of Electrical Engineering, The Netherlands b.gokberk@utwente.nl Abstract. In this work, we propose a fully automatic pose and expres- sion invariant part-based 3D face recognition system. The proposed sys- tem is based on pose correction and curvature-based nose segmentation. Since the nose is the most stable part of the face, it is largely invariant under expressions. For this reason, we have concentrated on locating the nose tip and segmenting the nose. Furthermore, the nose direction is uti- lized to correct pose variations. We try both one-to-all and Average Nose Model-based methodologies for registration. Our results show that the utilization of anatomically-cropped nose region increases the recognition accuracy up to 94.10 per cent for frontal facial expressions and 79.41 per cent for pose variations in the Bosphorus 2D/3D face database. 1 Introduction An effective face recognition system has to be fully automatic and robust enough for real life conditions where illumination, rotation, and expression variations are present. Although state-of-the-art 3D face recognition systems provide identifica- tion rates up to 99 per cent, they are mostly constrained by certain assumptions such as slight expression and pose variations [1]. Overcoming the problems due to extreme expression and pose changes still remains as a challenging problem. A recently proposed approach for expression invariant 3D face recognition is a part-based system [2]. Part-based approach is useful to alleviate the pose, facial expression and partial occlusion effects on the recognition performance. To deal with these effects, the entire 3D model of the face is split into several regions and these regions are processed separately. Aly¨ uz et al. [2] split the face into patches, and carry out an exhaustive search of all possible combinations of these surface to find the best subset of all the patches around the whole facial surface. Moreno et al. [3] segment the 3D facial surface using signs of mean, Gaussian curvatures and several three dimensional descriptors. Cook et al. [4] use Log- Gabor Templates on range images to deal with occlusions, distortions and facial expressions. Recently, Kakadiaris et al. proposed matching of spin images prior to ICP for alignment. They used wavelet analysis for feature extraction and obtained good recognition results in the presence of facial expressions [5]. In [6], Chang et al. use multiple regions selected around the nose area which have the M. Tistarelli and M.S. Nixon (Eds.): ICB 2009, LNCS 5558, pp. 309–318, 2009. c  Springer-Verlag Berlin Heidelberg 2009