非表示:
キーワード:
-
要旨:
We present two different techniques and applications that are based on the 3D
Morphable Face Model.
In the first part of this thesis, we develop a new top-down approach to 3D data
analysis by fitting a 3D Morphable Face Model to 3D scans of faces. The
algorithm is specifically designed for scans which were recorded in a
perspective projection. In an analysis-by-synthesis approach, shape, texture,
pose and illumination are optimized simultaneously. Starting from raw 3D scans,
the algorithm determines a PCA-based representation which fits the scan best.
Also, fragmentary surfaces are completed and correspondence to a reference face
of the morphable model is established. Simultaneously, illumination conditions
are estimated in an explicit simulation that involves specular and diffuse
components. The effects of lighting and shading are removed to obtain an
illumination corrected texture which stores the diffuse reflectance in each
point of the facial surface. We use the algorithm as a core component in 3D
face recognition on a subset of the FRGC database of scans.
In part two of this thesis we explore the growth of 3D faces, represented in a
3D Morphable Face Model. Assuming that 3D faces follow curved trajectories in
face space as they age, we present a novel algorithm that computes individual
aging trajectories for given faces. From a database of 3D scans of teenagers
and adults, we learn an non-linear function, that assigns an age to each face
vector using support vector regression. Computing the gradient of this function
leads us to trajectories that describe the direction of growth in face space.
Starting from photographs of faces we apply the aging prediction to images of
faces by reconstructing a 3D model from the input image and applying the aging
transformation on both shape and texture. The resulting face model is rendered
back into the same image or into images of other individuals at the appropriate
ages, for example images of older children. Also we may compute a variety of
possible appearances by changing attributes such as haircut, cloth or
background. Among other applications, our system can help to find missing
children.
