This research present a method of 2D and 3D segmentation based on the Enhanced Hoshen-Kopelman algorithm and its extension to non-lattice structure. The main feature of this method is to combine a merging strategy of a region growing algorithm with the multiple labeling technique of the EHK algorithm for regular and non-regular lattice. An efficient reconstruction algorithm is then applied to the set of edge points for the obtained segmented regions. The latest uses 3D Delaunay weighted triangulation. The combination of these already known algorithms makes the proposed approach very fast, efficient and appropriate to volumetric segmentation and particularly the anatomical structures in MRI and CTScan images.