Centroid estimation based on MSER detection and Gaussian Mixture Model

This paper proposes a sub-pixel centroid estimation algorithm for star field image recorded by CCD sensor. To handle the connected as well as partially overlapped star spots in a star region, we involve the the MSER (Maximally Stable Extremal Region) detector to discover the star spots and related characteristics of latent star spots in a star region. After extracting the star regions, we will estimate the locations of centroids in each star region simultaneously using the GMM (Gaussian Mixture Model) and Expectation-maximization (EM) algorithm, instead of segment each star spots in a star region. Using our algorithm, overlapped star spots can be discovered easily and their centroids can also be estimated effectively. Experiments on simulation data sets demonstrate effectiveness and robustness of this algorithm.