基于GPU的实时景深模拟  

Real-time depth-of-field simulation on GPU

作　　者：黄蓝枭[1] 卢光辉[1] 何明耘[1] 蔡洪斌[1] 

机构地区：[1]电子科技大学数字媒体技术研究所,成都610054

出　　处：《计算机应用研究》2008年第10期3045-3047,共3页Application Research of Computers

基　　金：国家"863"计划资助项目(2006AA01Z335)

摘　　要：在研究透镜成像模型与针孔成像模型的基础上,在计算机三维场景成像中实时地模拟出景深效果。算法利用了MRT(multiple render targets)技术及GPU的可编程性,在渲染时将场景存为纹理,并输出了像素的深度值和模糊因子,利用模糊因子计算每像素模糊圈大小并对图像进行滤波,最终在模糊圈内对清晰图像和模糊图像进行融合,模拟出景深效果。以一个场景的景深模拟为例,展示了不同聚焦时的景深效果。This paper proposed an efficient and real-time algorithm for the simulation of camera imaging in the three-dimensional scene based on both ideal and real camera models. The algorithm took extensive use of the multiple render targets （MRT） technology and graphics processing unit（GPU） for programmable performance. It outputed the image of the three-dimensional scene and stored as a texture, and also outputed each pixel＇ s depth and blurriness information and also stored as texture. ＂At last, filtered the image, calculated the size of confusion（CoC） for each pixel by the blurriness, and used CoC to blend between the original（clear） image and blurred image. Our experiments on simulation of depth-of-field in a geome-try scene demonstrate various effects with different parameters.

关 键 词：景深 虚拟现实 实时性 可编程图形硬件 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]