When compositing computer generated images and photographs or video images, interactions between real and virtual worlds must be taken into account in order to have a good visual result. These interactions are of three kinds: occluding object determination, specular inter-reflection and shadow computation. We first determine which real objects need to be geometrically and/or photometrically modelled for the three sets of interactions. Indeed, not all real surfaces need to be kept for the simulation and rendering steps. The unuseful real surfaces would just increase the rendering time. A rough photometric model can be sufficient for the surface properties, but colour computation has to be made in a particular colour model. Then for each kind of interactions, we take the real surfaces into account in a view-independent illumination step or in the rendering step.