Object Identification by Marked Point Process

In this paper, we propose an algorithm for the identification of objects from a noisy and cluttered background in video sequences. Our algorithm is based on the marked point process (MPP) framework, which provides a useful tool for integrating object spatial information into the identification process. The maximum a posteriori (MAP) estimation of a set of points corresponding to the centroids of objects observed in the image is obtained via a Markov chain Monte Carlo algorithm. The optimal solution, in terms of the MAP principle, is computed with respect to all objects in the scene, rather than single objects. The algorithm is applied to real data: intravital microscopic rolling leukocyte video datasets. A quantitative study of our approach demonstrates that the proposed approach can serve as a fully automated substitute to the tedious manual rolling leukocyte detection process