Real-Time Lens Blur Effects and Focus Control: Supplementary Material

Both mentioned game techniques use non-physical lens models, which makes a fair comparison difficult. In Figure 1, we tried to match the blur level of the single-image approaches. Due to the comparably small blur, our solution only needs 32 rays and 2 layers to produce a high-quality output. Nonetheless, even with these settings, our method is still slower than a filtering solution. Our approach needs 14 ms per image (of which 3 ms are used by the raytracing process) which is approximately twice as slow as [Filion and McNaughton 2008] with 6 ms (5.4 ms for rendering, 0.6 for postprocessing). Nevertheless, artifacts appear in the postprocessing methods, even for such a small blur radius. Our solution delivers a convincing outcome and can treat larger blur kernels gracefully. The same findings hold when comparing our method to another recent mipmap-based solution [Lee et al. 2009b]. This method shares the idea of the former approaches, but matches a physical lens behavior more closely.