This paper discuss a method suitable for inpainting
both large scale geometric structures and more
stochastic texture components. Image inpainting
concerns the problem of reconstructing the intensity
contents inside regions of missing data. Common
techniques for solving this problem are methods
based on variational calculus and based on
statistical methods. Variationalmethods are good at
reconstructing large scale geometric structures but
have a tendency to smooth away texture. On the
contrary statistical methods can reproduce texture
faithfully but fails to reconstruct large scale
structures. In this paper we use the well-known
FRAME (Filters, Random Fields and Maximum Entropy)
for inpainting. We introduce a temperature term in
the learned FRAME Gibbs distribution. By sampling
using different temperature in the FRAME Gibbs
distribution, different contents of the image are
reconstructed. We propose a two step method for
inpainting using FRAME. First the geometric
structure of the image is reconstructed by sampling
from a cooled Gibbs distribution, then the
stochastic component is reconstructed by sample
froma heated Gibbs distribution. Both steps in the
reconstruction process are necessary, and contribute
in two very different ways to the appearance of the
reconstruction.