Version 1.4.4

• Fixed k-means random initialization error that caused it to drop some
  clusters after the first iteration.

Version 1.4.3

• Fixed some k-means parser errors.
• K-means++ deterministic (det++) now works using a pivoting approach.

Version 1.4.2

Euclidean distance (PNorm metric with p = 2.0, which is the default value) is now much faster.

Despite the note that this commit is the same as 1.4.1, this is the correct version. I forgot to push the repository before creating this release.

Version 1.4.1

• Changed 'det++' initialization.
• Added weighing factor to LeaderFollower technique.

Version 1.4.0

• Correlation has been changed to PearsonsCorrelation.
• SpearmansCorrelation added as metric.
• LeaderFollower may use the following metrics: Cosine, PearsonsCorrelation and SpearmansCorrelation. Its usage has been simplified.

Version 1.3.5

• Added cosine metric.
• The cluster object now stores the mean spatial centroid.
• The deterministic k-means++ initialization allows the user to set the first centroid.
• Modified Leader-Follower. It now sorts the voxels by peak time and amplitude before inspecting them. Also, returns the voxel with the smallest Euclidean distance to the centroid.

Version 1.3.4

• Changed LeaderFollower behavior.

Version 1.3.3

• Fixed the iterator. Again. Once and for all.
• Fixed a bug in the RealMatrix2IJ method that caused it to shift slices.

Version 1.3.2

• A new bug was introduced in the last update -- first voxel was always returned by the image iterator. Fixed.

Version 1.3.1

• Implements PlugIn interface instead of PlugInFilter.
• Solved bug in iterator (not returning first voxel) that caused PCA / SVD /ICA to crash on unmasked images.