Stochastic mean-shift clustering
This is an incremental improvement for clustering algorithms, potentially benefiting data analysis in fields like pattern recognition or image segmentation.
The paper tackles the problem of improving mean-shift clustering by introducing a stochastic version where data points collectively converge to distribution modes, and it found that this stochastic approach outperformed the deterministic version in most cases on synthesized 2D and 3D Gaussian data in terms of cluster purity and class data purity.
In this paper we presented a stochastic version mean-shift clustering algorithm. In the stochastic version the data points "climb" to the modes of the distribution collectively, while in the deterministic mean-shift, each datum "climbs" individually, while all other data points remains in their original coordinates. Stochastic version of the mean-shift clustering is comparison with a standard (deterministic) mean-shift clustering on synthesized 2- and 3-dimensional data distributed between several Gaussian component. The comparison performed in terms of cluster purity and class data purity. It was found the the stochastic mean-shift clustering outperformed in most of the cases the deterministic mean-shift.