User:Jadrian Miles/Chi-squared clustering
Jump to navigation
Jump to search
This paper brings together two existing projects with two new contributions:
- Estimating the distribution of noisy tractography streamlines around a known "true" curve is described in my ISMRM 2012 submission.
- Curve-to-curve similarity is described in my NIH talk from 2009. Code for this is in ~/gfx/brain_imaging/curveClustering. The Matlab code was a prototype; the cpp code is a full implementation.
- curveDistRowClient computes certain rows of the pairwise distance matrix. This stuff should be rewritten around the curveCollection library and checked into $G.
- The mean endpoint distance is probably a more appropriate and (computationally and conceptually) lighter-weight measure to use.
- Initialization and merging of curve bundles is new.
- Optimization of the chi-squared reconstruction error is new.