586 C HAP TER 11. IMAGE-BASED LIGHTING Ambient occlusion most accurately approximates the correct lighting solution when the lighting environment is relatively diffuse. In this case, the homogeneous environment used to compute the occlusion is a close approximation to the envi- ronment desired to light the scene. Ambient occlusion is not designed to simulate light from environments that include concentrated light sources, as the directional detail of the environment is lost in the diffuse convolution process. For IBL environ- ments that do have concentrated light sources, an effective way of handling them is to simulate them as direct light sources (as described in Section 11.5.1), delete them from the IBL environment, and use a diffuse convolution of the modified IBL environment to multiply the ambient occlusion map. Although computing ambient occlusion maps requires sending out a multitude of rays to the lighting environment, the number of rays that need to be sent is mini- mized because the environment has minimal variance, which alleviates the sampling problem. Also, the ambient occlusion map is solely a function of the object geometry and is independent of the lighting environment. Because of this, the technique can be used to render an object with different lighting environments while performing the ambient occlusion calculation map only once. This makes real-time implemen- tations very fast, especially for rotating lighting environments for which perform- ing additional diffuse convolutions is also unnecessary. In addition, the technique allows for relighting effects to be performed inside a standard compositing system. For example, the convolved light probe image can be manually edited and a relit ver- sion of the scene can be created quickly using the preexisting normals and ambient occlusion map without rerendering. 11.8 IMAGE-BASED LIGHTING FOR REAL OBJECTS AND PEOPLE The IBL techniques described so far are useful for lighting synthetic objects and scenes. It is easy to imagine uses for a process that could illuminate real scenes, objects, and people with IBL environments. To do this, one could attempt to build a virtual model of the desired subject's geometry and reflectance, and then illuminate the model using the IBL techniques already presented. However, creating photoreal models of the geometry and reflectance of objects (and particularly people) is a difficult process, and a more direct route would be desirable. In fact, there is a