Segmentation of clusters by template rotation expectation maximization

To solve the task of segmenting clusters of nearly identical objects we here present the template rotation expectation maximization (TREM) approach which is based on a generative model. We explore both a non-linear optimization approach for maximizing the loglikelihood and a modification of the standard expectation maximization (EM) algorithm. The non-linear approach is strict template matching , while the linear TREM allows for a more deformable model. As benchmarking we compare TREM with standard EM for a two dimensional Gaussian mixture model (GMM) as well as direct maximization of the log-likelihood using non-linear optimization. We find that the linear algorithms, TREM and standard GMM, are faster than the non-linear algorithms without any loss of segmentation accuracy. In the comparison between the linear models we find an edge for the linear TREM on our data as it better approximates the orientation and shape of the objects than the standard GMM.