Traditional mapping techniques rely on metric properties, which represent indoor information with specific geometric characteristics. This fact highly differs from the way in which people interpret their surroundings.By geometrically segmenting occupancy grid maps into rooms, robots are brought closer to the way in which we understand indoor environments. In this work, Voronoi diagrams are proposed as the main tool to locate map partitions. As a novelty, they are extracted from free and occupied spaces to analyze their shape. This allows to locate narrow passages on free zones which coincide with protruding parts on occupied zones,indicating a nearby door. An additional advantage is the use of a varying threshold that depends on the map structure. This dynamic value can adjust to multiple scenarios, avoiding the use of a fixed threshold that can not be generalized. Experiments have been conducted in multiple maps, showing the potential of the propose method.