dbscan clustering of xy points only returns outliers

I have an array of xy points and just want a way to determine the number of clusters. I have many sets of xy points, and all of them are approximately the same in terms of values, so I would think epsilon doesn't need to change too much.

I have tried

idx=dbscan(centers,1,3);

like they did in the example, but it only returns a vector of -1, which corresponds to outliers. I thought decreasing the epsilon would give more clusters. Using kmeans with 2 clusters gives me the below result, but I would like to separate them even more. I can see both the red and the blue clusters could each have two more. Do I just not have enough points?

Answer 1

[the cyclist]

You actually need to increase epsilon, because you need a larger search radius to identify neighborhood points.

load centers
idx = dbscan(centers,50,3);
gscatter(centers(:,1),centers(:,2),idx)