Research And Implementation Of Fast Depth-of-field Rendering

When taking photos，we can fnd a phenomena that objects in the focus regionare clear while background and foreground are blurred. The phenomena is Depth offeld (DoF). Precisely, DoF is a depth range outside of which objects are blurred. DoFhas been widely used in photography and flms by artists and is expected to be appliedto3D video games and virtualreality applications. Hence, fast DoFrendering methodscan be used in interactive artistic tools to realize real-time preview and in3D game torealize real-time realistic rendering.Due to lens and infnite aperture, a point of object in foreground or backgroundproject to image plane and form a circle（circle of confusion）. According to geomet-rical optics，size of circle of confusion can be calculated from depth of each point.This relationship is theoretical foundation of fast DoF rendering methods.In this paper, we survey classical methods based on the objects space as well asfast DoF rendering methods(post-processing method: using sharp scene image ren-dered by pinhole camera model and depth map to generate fnal image with DoF efec-t).We analyze previous methods in several aspects including quality of fnal image andperformance. Moreover,we point out problem in these methods and give the reasonscausing them. Then, based on the achievement of analyzing previous methods，wesummarize kinds of artifacts（intensity leakage，depth discontinuity，lack of partialocclusion and depth discretization），point out the reason causing these artifacts andpropose methods to remove these based on the previous methods.WeproposelayeredDepth-of-FieldrenderingmethodUsingcolorspreading. Themain idea of us is using depth peeling to get hidden surface information and using col-or spreading to decompose all information.The method involves two main stages.The frst stage is preprocessing, including depth peeling,real depth and CoC calculating inturn.The second stage involves scene decomposing and blending. The key contributionof this paper is a layered DoF efect framework that gives results with high quality andperformance by spreading the fragments to build scene layers, leading to lower mem-ory cost compared with those that store the layers for blurring. And a modifed depthpeeling is proposed to reduce the number of depth layers.Finally, we implement our method in GPU and optimize it to make the best useof the GPU. The accumulation bufer method、gathering method and another layeredmethod are used to compare with our method. The result shows that our method canrender DoF efect with high quality and performance.