2D + 3D approach in face verification algorithm

• Main Author: Hendinata, Joshua
• Other Authors: Li Fang
• Format: Final Year Project
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/63472
• Institution: Nanyang Technological University

Typical face recognition system usually use only 2D face image captured from a single camera. This method is preferred compared to 3D face recognition which, although offers a generally higher accuracy rate, was expensive due to the use a high-end range camera or laser scanner for its acquisition process. However, due to its nature, 2D face recognition is often prone to spoof attack such as photo and video attack. This project aimed to combine best of both worlds by using ubiquitous gadget such as laptop and mobile camera to create a low-cost and robust, yet accurate face verification system. The system implements a stereo camera setup to calculate face disparity map which capture face depth information. Semi-Global Block Matching (SGBM) was implemented as stereo matching algorithm and Local Binary Pattern Histogram (LBPH) as its main face recognition algorithm. The experiment conducted with author's self-sourced database, which consists of 100 images from 10 subjects, yield a significant improvement in accuracy compared to 2D face recognition alone.