Face recognition using 3D line edge map based on multiview stereo images

Abstract

In this thesis, we propose new face recognition system using 3D Line Edge Map as combining 3D information and the conventional Line Edge Map algorithm. 3D information is obtained from a stereo camera and estimated only on the edge to acquire reliable values. Images in various views are in the database to treat various viewing position of input images with ease. Poses of input images are estimated using view-based PCA. As a similarity measure of 3D Line Edge Map, weighted partial and spatially well-localized line segment Hausdorff Distance (WPSLHD) is proposed. WPSLHD is a measure which is robust to expression changes, facial occlusions, and depth outliers. We determine the weight of each segment in a reference Line Edge Map by the position variance and the orientation variance of corresponding line segments. Also, we remove depth outliers and outliers by occlusions searching corresponding line segments among the spatially well-localized segments. The proposed face recognition system performances are evaluated by several experiments under facial expression changes and facial occlusions, and in intermediate views. The experimental results show that the proposed 3D Line Edge Map algorithm and its similarity measure WPSLHD achieve higher recognition rate than the conventional Line Edge Map algorithm and LHD.