We apply to edge detection a recently introduced method for computing geome
tric structures in a digital image, without any a priori information. Accor
ding to a basic principle of perception due to Helmholtz, an observed geome
tric structure is perceptually "meaningful" if its number of occurences wou
ld be very small in a random situation: in this context, geometric structur
es are characterized as large deviations from randomness. This leads us to
define and compute edges and boundaries (closed edges) in an image by a par
ameter-free method. Maximal detectable boundaries and edges are defined, co
mputed, and the results compared with the ones obtained by classical algori
thms.