This paper presents a method for the automatic extraction of the structure of wastewater networks from geographical maps, as well as their representation in the form of graphs. The approach consists first in detecting the different important elementary elements composing a wastewater network, such as the manholes, their identifiers (using optical character recognition, OCR), and the wastewater pipes that connect them. Detecting these elementary elements is a first difficult problem, despite many existing tools, mainly due to the quality of the wastewater network maps used. However, the challenge addressed by this paper is how to select the relevant detected elementary elements, and put them together to finally extract the wastewater network. One of the main contributions of this paper is to propose an efficient algorithm to solve these selection and assignment problems (e.g., manhole identifiers to manholes represented by circles). We also deal with the situation of isolated nodes to have the most connected clusters possible. The experimental results conducted on real maps data show very interesting results despite the low quality of the maps.