DBScan Clustering for test dataset from learnt trained model

From the code, i can see that DBScan maps every instance in training set to a cluster based on density. But wanted to see how given a new data/test set it can use the clustering model to map to existing cluster? In KMeans atleast we have centroids that are learnt and we use distance measure to find closest distance of data to the learnt clusters, how does DBScan do this?
I wanted to compare EM, DBScan and KMeans on a given problem(train, test) and see how they map.

Thanks
John