基于可编程图形处理器的实时景深模拟

一幅具有景深效果的图，其显著特征是图像部分清晰部分模糊。基于这一特点，在景深的计算机仿真中，首先由计算机对三维场景成像，并将该场景存储为纹理，此纹理为清晰的场景纹理。成像时通过顶点编程，于α通道中存储了物体距离聚焦面的深度信息。然后对场景纹理进行多次均值滤波生成模糊的场景纹理，最后以清晰场景纹理的α值为插值系数将两幅纹理进行融合而模拟出景深效果。算法充分利用了图形处理器（GPU）的多纹理技术和可编程性，将运算从CPU转移至GPU，很好地仿真了景深效果，且满足实时性要求，可应用于虚拟现实系统。

Programmable GPU Based Real-time Simulation of Depth of Field

The prominent characteristic of an image with depth of field(DOF) effect is that part of the image is clear but the other part is blurry.From this point,in the computer simulation of DOF,an image of the three-dimensional scene was produed and it was stored as a texture which was called as clear scene-texture.In the process of imaging,the information of the depth of objects was stored in the alpha channel by vertex program.Then the clear scene-texture was filtered averagely several times and another blurry scene-texture was got.At last,the alpha value stored in the clear scene-texture was taken as the coefficient,which used it to blend these two textures to achieve the DOF effect.The algorithm takes extensive use of the Graphics Processing Unit(GPU) for its multi-texture and programmable performance,and thus transfers the considerable calculation from the CPU to the GPU.The algorithm can well simulate the real-time DOF effect,and can be applied in VR system.