GSoC 2011

Vastly increase the number of algorithms covered by the SciKit, by implementing several prominent algorithms (probabilistic hough, radon transform, etc.) from papers. Also incorporate code from the Broad Institute's CellProfiler team.

Implement a new backend infrastructure so that the same results may be obtained using either NumPy, OpenCL, Theano, etc.

Implement optimised core routines, such as fast 2D convolution. This should be as fast as CvFilter2D (which is faster than ndimage, for example).