Expression-invariant face recognition for small class problem

Face recognition has been recognized as most simple and non-intrusive technology that can be applied in many places without possible hazardous problems. However, there are still many unsolved face recognition problems (especially when real-time identification is required) due to different facial expressions (deformations), poses, illumination or occlusions etc. Although many researchers have been working to find solutions for these problems so as to achieve robustness under various circumstances, little research has been done on facial deformation problem. The objective of this research was to determine if a face recognition system could be invariant to facial deformations. The database used here was called Japanese Female Facial Expression (JAFFE) database. We chose 18 geometric facial attributes from 80 landmark fiducial points of a face for facial features. We used Principle Component Analysis (PCA) and Fisher's Linear Analysis (FLD) methods of subspace analysis for dimensionality reduction. We then used Multi-layered Perception (MLP) and Radial Basis Function (RBF) neural networks for decision-making. The experimental results were compared with results from those classifiers without using subspace analysis. The results used the FLD subspace analysis method showed to achieve better classification performances. These findings indicate that face recognition may be robust to facial expression changes.