Many tasks in image processing, computer vision and computer
graphics require to interpolate images in order to estimate
data at locations where no values are specified.
In this talk a common framework for image interpolation
and regularisation is investigated. This framework is based
on partial differential equations (PDEs) and allows
rotationally invariant models. It can be used for a large
range of applications including resampling, scattered data
interpolation and image inpainting. Experiments illustrate
that it outperforms interpolants with radial basis functions,
it allows discontinuity-preserving interpolation with no
additional oscillations, and it respects a maximum-minimum
principle. These ideas are not restricted to scalar- or
vector-valued images, they can also be applied to matrix-valued
data sets arising e.g. from diffusion tensor magnetic resonance
imaging. In this case the interpolated matrix field preserves
the semidefinite positiveness of the initial data.