Multi-layer screen-space ambient occlusion using hybrid sampling

Ambient occlusion is a shading method which approximates the global illumination by taking into account attenuation of light due to occlusion[Christensen 2002; Sergey et al. 1998; Andrei et al. 2003]. Screen-space ambient occlusion is a scene-independent improvement which performs the visibility test using depth buffer on GPU in real time. Shanmugam et al. [Shanmugam et al. 2007] proposes a real-time multi-pass hardware assisted ambient occlusion method. The algorithm randomly samples the normal/depth buffer and reconstructs the surface using spheres that can be projected to screen, but has the overocclusion issue. Bavoil et al. [Bavoil et al. 2008] approximates the ambient occlusion illumination using a image-space horizon-based method. However, the scenes will also be overoccluded due to the inconsideration of the layers behind the frontmost one. Moreover, the rough approximation of the geometry occluder makes the shadows lack of rich details. In this paper we propose a new solution to approximate ambient occlusion using a hybrid sampling method of both above methods. The surfaces can be more accurately depicted using spheres and the geometry occluder under the horizon plane can be discarded to address the overocclusion issue. In addition, a multi-layer method is utilized to future alleviate overocclusion using the depth information of all the layers. Experiments show that our algorithm can generate more realistic results for complex scenes with rich details, with comparable performance.