Face Recognition Based on Multiple Region Features Jiaming Li, Geoff Poulton, Ying Guo, Rong-Yu Qiao CSIRO Telecommunications & Industrial Physics Australia Tel : 612 9372 4104, Fax : 612 9372 4411, Email : jiaming.li@csiro.au Abstract. For face recognition, face feature selection is an important step. Better features should result in better performance. This paper describes a robust face recognition algorithm using multiple face region features selected by the AdaBoost algorithm. In conventional face recognition algorithms, the face region is dealt with as a whole. In this paper we show that dividing a face into a number of sub-regions can improve face recognition performance. We use conventional AdaBoost with a weak learner based on multiple region orthogonal component principal component analysis (OCPCA) features. The regions are selected areas of the face (such as eye, mouth, nose etc.). The AdaBoost algorithm generates a strong classifier from the combination of these region features. Experiments have been done to evaluate the performance on the CMU Pose Illumination Expression (PIE) databases. Performance comparisons between single region OCPCA, our multiple region OCPCA, and published results from Visionics’ FaceIt are given. Significant performance improvement is demonstrated using multiple facial region OCPCA features. Keywords: face recognition, multiple region features, AdaBoost, Orthogonal Component PCA. 1 Introduction In addition to popular biometric measures such as fingerprint and iris scan, the human face is another common biometric that can be used for automatic person recognition. CSIRO has developed a real time face capture and recognition system, which can automatically capture and recognise face [1]. The recognition of faces is done by finding the closest match of a newly presented face to all faces known to the system. Popular face recognition methods include Principal Component Analysis (PCA) [2][3], Independent Component Analysis (ICA) [4], Neural Networks [5], etc. [6][7]. Generally, face recognition algorithms treat the face as one entity when performing face recognition. However, the face can be divided into a number of regions and separate features derived for each region. It would be show that such division can improve classification performance. Note that regions may overlap 69 Proc. VIIth Digital Image Computing: Techniques and Applications, Sun C., Talbot H., Ourselin S. and Adriaansen T. (Eds.), 10-12 Dec. 2003, Sydney