K-Means for noise-insensitive multi-dimensional feature learning

Many measurement modalities which perform imaging by probing an object pixel-by-pixel, such as via Photoacoustic Microscopy, produce a multi-dimensional feature (typically time-domain signal) at each pixel. In principle, the many degrees of freedom in signal would admit possibility significant multi-modal information being implicitly present, much more than single scalar “brightness”, regarding underlying targets observed. However, measured is neither weighted-sum basis functions (such principal components) nor one set prototypes (K-means), has motivated novel clustering method proposed here. Signals are clustered based on their shape, but not amplitude, angular distance, and centroids calculated direction maximal intra-cluster variance, resulting algorithm capable learning (signal shapes) that related to underlying, albeit unknown, target characteristics scalable noise-robust manner.