Added support for square filters.

[anweidong]

Use integral image to compute the encodings and costs.

To use the square filters google = imresize(imread('google.jpg'),0.5); mat = AMAT(google, 'shape', 'square');

Now the costs is using mean square errors. There is also an implementation for the heuristic costs (Line 699-713) but some issues along the boundaries remains so that part of the code is commented out.

The implementation is ~10s faster than the disk one with mean square errors and ~7s faster with the heuristic costs. However the result is not as good as the disk implementation especially for the reconstruction after grouping and simplifying.

Looked into the computeReconstruction and it is not necessary to change.

[anweidong]

• Heuristics costs added
• bug fixed along the boundaries

Result for the google logo

[anweidong]

Added the oriented squares filters. Added implementation for multiple kinds of filters.

google = imresize(imread('google.jpg'),0.5); mat = AMAT(google, 'shape', 'mixed');

This implementation is not working so well for setCover, but the visualization of encoding and costs makes sense now.

One scenario now works fine. Set (line 10) rotations = [] i.e. the disk and square filters without rotation. Debugging what still needs to be changed for rotated squares.

[anweidong]

Bug fixed for the rotated square filters. Run the following code:

google = imresize(imread('google.jpg'),0.5); mat = AMAT(google, 'shape', 'mixed');

Result for the google logo image.

[anweidong]

Added implementation for combined filter by rotating the image. Run the following mat = AMAT(google, 'shape', 'mixed2');