We propose an edge detector based on the selection of well/ncontrasted pieces of level lines, following the proposal of/nDesolneux-Moisan-Morel (DMM) [1]. The DMM edge detector/nhas the problem of over-representation, that is, every/nedge is detected several times in slightly different positions./nIn this paper we propose two modifications of the original/nDMM edge detector in order to solve this problem. The first/nmodification is a post-processing of the output using a general/nmethod to select the best representative of a bundle of curves./nThe second modification is the use of Canny’s edge detector/ninstead of the norm of the gradient to build the statistics. The/ntwo modifications are independent and can be applied separately./nElementary reasoning and some experiments show/nthat the best results are obtained when both modifications are/napplied together.