I have seen many DBSCAN algorithm implemented using a formula to estimate the neighborhood radius (Eps) based on the given minimum points within a cluster (k).
[full code] http://toolz.googlecode.com/svn/trunk/CWT/dbscan.py
% Analytical calculation of rad if not given
function [Eps] = epsilon(x,k)
[m,n] = size(x);
Eps = ((prod(max(x)-min(x))*k*gamma(.5*n+1))/(m*sqrt(pi.^n))).^(1/n);
I have searched extensively to understand how this analytical formula was derived but been unsuccessful.