Ported vigranumpy to Python 3.5.
Added chunked arrays to store data larger than RAM as a collection of rectangular blocks.
Added vigra::ThreadPool and parallel_foreach() for portable algorithm parallelization based on std::thread.
Implemented parallel versions of Gaussian smoothing, Gaussian derivatives, connected components labeling, and union-find watersheds.
Added graph-based image analysis, e.g. agglomerative clustering
Included the callback mechanism described in Impossibly Fast C++ Delegates by Sergey Ryazanov (needed for agglomerative clustering).
Added many image registration functions.
Extended the collection of multi-dimensional distance transform algorithms by vectorial DT, boundary DT, and eccentricity transform.
Added skeletonizeImage(), nonLocalMean(), multi-dimensional integral images.
Added new 2D shape features based on skeletonization and the convex hull.
Additional arithmetic and algebraic functions for vigra::TinyVector.
Added vigra::CountingIterator.
Minor improvements and bug fixes in the code and documentation.
da99a7bd96