Advanced Multimedia Processing Lab -- Projects -- Portable 3D Faces for Field Identification of Suspects

            About AMP Lab        Projects        Downloads        Publications        People        Links

Project - Portable 3D Faces for Field Identification of Suspects

Top of this page

Motivation and Goal

  • Construct a 3D face model based on 2D images or video captured by low-resolution surveillance cameras.
  • Display a 3D face model with face images rendered at different poses, under various illumination conditions, and with different surroundings so as to ease suspect identification.
  • Manipulate the 3D face model to create different facial expressions, to simulate the aging process, to add or remove facial hairs, or to change hairstyles or other decorations.

Top of this page

  • Stage I:  Train generic (i.e., prior) face model using many examples:
  • 3D mesh model
    1. Collect face images from various poses



    2. Gather subjects' 3D information by using stereo vision techniques on above poses (or by using laser range scans)
    3. Build generic 3D mesh
      3D face mesh
  • 2D texture model
    1. Collect face image textures for both high and low resolutions


    2. Build generic model (e.g., Markov random field, MRF) for generating texture's missing high-frequency information.
                  
  • 3D mesh & 2D texture
Overlay generic 2D texture onto generic 3D mesh
3D face mesh

  • Stage II:  Synthesize face model from single subject's image(s)
  • Adapt generic 3D mesh using marked points on subject's low-resolution image
    Generic 3D Face MeshAdaptd 3D Face Mesh:
  • Enhance low-resolution 2D texture using trained model from Stage I (simulated)
  • Overlay enhanced 2D texture onto adapted 3D mesh
    3D model with adapted mesh and enhanced texture

Top of this page

Any suggestions or comments are welcome. Please send them to Todd Stephenson. 

Top of this page


This website is maintained by Todd Stephenson
Copyright 2005
Advanced Multimedia Processing Lab. All rights reserved.
Revised: August 23, 2005 .