Face recognition using 2D and disparity eigenface

Face recognition has been a popular research topic in computer vision. Numerous different face recognition techniques have been developed owing to the growing number of real world applications. This study presents a face recognition method based on the stereo matching and eigenface techniques under a stereovision system. The proposed method aims to improve the performance of the 2D eigenface system by adding 3D information. 3D information (called disparity face) was derived by taking two faces of a subject simultaneously in different positions, and then the two faces were further matched using a scanlined-based asynchronous Hopfield neural network. After deriving 2D and 3D faces, we applied principal component analysis (PCA) to both faces to extract effective features for recognition. An experiment was conducted acquiring the facial images of 100 individuals. Each subject provides three pairs of faces with different expressions for training and testing. At the finals, the performance of face recognition using 2D faces, disparity faces, and a combination of the two was evaluated and compared. The experimental results reveal that a 3-5% improvement in recognition rate is achieved by using the additional depth information.