Material Detail

Facial Image Analysis using Directional Statistics and Shape-from Shading

Facial Image Analysis using Directional Statistics and Shape-from Shading

This video was recorded at Workshop on Applications of Pattern Analysis (WAPA), Cumberland Lodge 2010. Although the recovery of facial shape using shape-from-shading is an appealing idea, it is frustrated by problems such as concave-convex ambiguities, variable albedo, self shadowing and non-Lambertian reflectance. As such the devil resides in the detail. In this talk I will show how these problems can be overcome by incorporating a statistical model for surface normal direction within the shape-from-shading process. The main contribution of the talk is to develop a representation of the distribution of surface normals using the equidistant azimuthal projection from cartography, which transforms a distribution of surface normal direction on a unit sphere to a distribution of points on a tangent plane. I will show how this model can be adapted to deal with shadowing by fitting the statistical model to image brightness data using robust statistics. I will also show to to adapt the process to deal with non-Lambertian reflectance, through fitting a reflectance model that can capture the behavour of both shiny and rough surfaces. Finally, I will show how the shape information delivered by the process can be used to perform face recognition and gender determination. This talk will provide a synopsis of recent work by Smith and Hancock (PAMI 07, IJCV09, IJCV 2010) and Wu, Smith and Hancock (IVC 2010)

Quality

  • User Rating
  • Comments
  • Learning Exercises
  • Bookmark Collections
  • Course ePortfolios
  • Accessibility Info

More about this material

Browse...

Disciplines with similar materials as Facial Image Analysis using Directional Statistics and Shape-from Shading

Comments

Log in to participate in the discussions or sign up if you are not already a MERLOT member.