DBSCAN with a larger neighbourhood

I am using sklearn.cluster.DBSCAN on my dataset. However, it tends to classify some of my data points as noise, even though they are not. However, if I increase eps even more, it will start merging unrelated clusters. I figured that my best attempt would be if I kept the clustering phase the same, but increased the allowed range for the "neighbour finding" phase. Is there such a possibility? My only other approach would be to build a kd-tree of all non-noise points, and for each noise point to look for the closest non-noise point and evaluate if they belong together. However, of course it would be better if this was working in a built-in way.