3D face recognition based on high-resolution 3D face modeling from frontal and profile views

  • Authors:
  • Lijun Yin;Matt T. Yourst

  • Affiliations:
  • State University of New York at Binghamton, Binghamton, NY;State University of New York at Binghamton, Binghamton, NY

  • Venue:
  • WBMA '03 Proceedings of the 2003 ACM SIGMM workshop on Biometrics methods and applications
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents a novel face recognition system which considers information from both frontal and profile view images and videos. In the system, we recover facial texture details by increasing the input image resolution, construct an accurate 3D face model from two views of a face, and explore both 3D shape and texture informations for an optimal match and identification based on a 3D face model database. Unlike many existing 3D face recognition systems where the 3D model is taken as a bridge for synthesizing textures of various poses from the viewing sphere, we explicitly use 3D geometric information to index the reference database in order to increase the matching accuracy. This work is the first step toward the development of a face recognition solution by exploring 3D context explicitly. The system consists of three major modules, including (1) 3D face model database creation based on two views' face images input; (2) query face model synthesis from two views' face video input; (3) matching between the database model and the query model using a hybrid method (i.e., shape and texture). The high resolution face model reconstruction is critical for success of the system. Five key components are developed: (1) Facial silhouette extraction; (2) facial texture detail reconstruction based on a novel algorithm: Hyper-resolution image enhancement; (3) feature detection from two views of a face; (4) face model instantiation by adapting the model to the resolution-increased input image; (5) 3D facial geometric information reconstruction using two views' models. The system has been tested using 60 subjects and has shown the correct match rate at 91.2%.