Abstract : Since the emergence of imaging techniques, MRI, CT, it is now possible to directly probe the geometrical organization of systems such as bone, cement, paper, glass, rocks. As the physical and mechanical properties depend on the geometrical organization, there is a scientific and industrial interest for the understanding of this relationship by using these imaging techniques. In this context, the purpose of this thesis is the development of a toolkit for digital image analysis of the material geometry, then the application of this toolkit in the study of the evolution of the pore space of cement paste. In part one, after a discussion on the choice of an imaging technique adapted to a material, we present the two imaging techniques selected, scanning electron microscopy and synchrotron tomography for the analysis of cement paste and the experimental protocol for sample preparation. In the second part, we propose a generic, efficient and simple methodology of segmentation. Segmentation is the transformation of a grey-level image to an labeled image where each label represents a phase of the material. Generic means that this methodology can be used for a wide range of materials and imaging techniques. Effective means that the segmented structure matches the real structure. Simple means that the calibration is easy. The implementation of the optimized algorithms associated with this methodology is done thanks to the theoretical conceptualization of the region growing. In the final part, we quantify the morphology and topology of the geometry of the material statistically. Then, we decompose a phase in term of elementary components along two agreements: one morphological and the other topological. Finally, we use the stereological information estimated on the 2D slice to reconstruct a 3D model larger than the representative elementary volume using the optimized algorithm of simulated annealing. The validation of the 3D reconstruction is performed by the comparison of properties of diffusive transport.