A separating algorithm for particle image

In the particle images, objects tend to touch, overlap with each other to form clusters. How to separate these clusters into single objects is a vital problem for the counting, feature extraction and analysis in the following process. Separating them by boundary tracking do not need an acute angle and/or the strong density contrast at the points where the objects touched, so it can get a better separating result. But sometimes when the principal axis direction of the cluster is neither horizontal nor vertical, it does not work. For this, a new separating algorithm is presented in the paper. At first, the algorithm calculates the principal axis direction of the cluster, if it is nearly horizontal or vertical then goes on, otherwise it rotates the cluster to make its principal axis direction be horizontal. And then the algorithm performs separating by boundary tracking. Experiments show that the algorithm has achieved a well separating result.