Final project of the Computational Intelligence Lab (ETH Zurich, Spring Semester 2012).
Alkis Gkotovos (alkisg@student.ethz.ch), Bo Li (libo@student.ethz.ch) and Alexey Sizov (asizov@student.ethz.ch)
A method for image inpainting combining Gaussian Process regression, intensity propagation, and sparse coding via an overcomplete dictionary.
The directory contains the files used by our combined method, as well as
the files of the baseline methods used for comparison. Note that the
main function for all methods is called inPainting.m
and all of them
adhere to the same interface.
The files of our combined method reside under the final/
subdirectory.
Included are also the files covMaterniso.m
, meanConst.m
, likGauss.m
,
infExact.m
, solve_chol
, and sq_dist.m
from the GPML toolbox
(http://www.gaussianprocess.org/gpml/code/matlab/doc/) that are used for
doing Gaussian process inference.
The GP baseline can be run using the code of the combined method, after
setting the flags use_dict
and use_ip
in file inPainting.m
to false
.
The files of the dictionary baseline method can be found under the
baselines/dict/
subdirectory.
The files of the SVD baseline method can be found under the baselines/svd/
subdirectory.
For evaluating any of the methods on our image test set, the following steps should be taken:
- Execute in Matlab the
startup.m
script found at the top directory. - Change the working directory to one of the methods, i.e. to
final/
for running the combined method or the GP baseline, tobaselines/dictionary/
for running the dictionary baseline, or tobaselines/svd/
for running the svd baseline. - Execute in Matlab the
EvaluateInpainting.m
script, which should have already been added to the path after the first step, to run the corresponding method on all test images available. Providing atrue
argument toEvaluateInpainting
will additionally print the reconstruction results as they are computed.
As an example, to evaluate our combined method and display the resulting reconstructed images, use the following commands in Matlab:
> startup
> cd final
> EvaluateInpainting(true)