Scatterometry has become an essential method of characterization during the fabrication process of nanostructures. This nondestructive technique based on diffracted light analysis is composed of two steps: the measurement of the optical signatures and the treatment to reconstruct the profile of the periodic structure. The artificial neural network has proved its effectiveness in solving the inverse scattering problem. Usually the scatterometry characterization process requires a previous geometrical profile shape to be defined. This study proposes a method for the profile reconstruction of any geometrical shape, not necessarily one that is included in the initial model. For example, this could include the detection or the identification of an incorrect profile shape in a lithography or etching process. This so-called weighting profile approach consists in combining several results of basic characterizations for the reconstruction of the real profile shape. In this study, the feasibility of this method is demonstrated on theoretical samples to satisfy the requirements of scatterometry. Finally, the weighting profile method is compared with a classical scatterometry method involving a generic profile shape.