Face Recognition Based on a Two-View Projective Transformation

• Main Authors: Chen Hui Kuo, 郭振輝
• Other Authors: J. D. Lee
• Format: Others
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/26869402648053870963

博士 === 長庚大學 === 電機工程學系 === 100 === In this dissertation, we propose two novel face recognition algorithms, one is cascade classifiers for face recognition undertaken by coarse-to-fine strategy using multiple samples per subject, and the other is face recognition based on a two-view projective transformation using one sample per subject. Face recognition scheme based on multi-stages classifier, which includes methods of support vector machine (SVM), Eigenface, and two-view projective transformation, is proposed in this dissertation. The whole decision process is undertaken by cascade coarse-to-fine stages. The first and second stages are SVM with one-against-all (OAA) and one-against-one (OAO) method picks out two classes with the least variations to the testing images. From the selected two classes, the third stage with Eigenface method decides the priority of images for a fine match with testing images at final stage with two-view projective transformation method. A fine class with greatest geometric similarity to testing images is thus produced at final stage. This multi-stage face recognition system has been tested on Olivetti Research Laboratory (ORL), Yale and Institute of Information Science (IIS) databases, and the experimental results give evidence that the proposed approach is superior to the previous approaches based on the single classifier in recognition accuracy. The drawback of SVM can’t use one sample per subject to build the recognition model. In order to solve this problem, we propose a novel face recognition algorithm based on two-view projective transformation, called the robust estimation system (RES). Our approach adopts both local and global information for robust estimation. We utilize the original images from the ORL and Yale databases for performance evaluation. The images of FERET database are pre-processed to extract the face region and execute the affine transformation. We roughly divide the face images into the four block images that are most significant for a face: left eye, right eye, nose, and mouth. The feature used here are magnitudes of first-order gradients. While conducting the classification stage, local features are putatively matched before the processing or the global RANSAC robust estimation features, with the aim of identifying the fundamental matrix between two matched face images. Finally, similarity scores are calculated, and the candidate awarded the highest score is designated the correct subject. Experiments were implemented using the FERET, ORL and Yale databases to demonstrate the efficiency of the proposed method. The experimental results show that our algorithm greatly improves recognition performance compared to existing methods.