Is there any possibilitie to create virtual shallow DOF with filters? I don´t refer to 3D applications but to real world footage: For example, take 2/3" camera footage and increase the blur of already blurred areas in post until you get a 35mm look. Would that bie possible, at least in certain circumstances?
I have not seen anyone do a good job at it with a pure computer vision / image processing basis except with specially callibrated multiple simultaneuous views (eg the Microsoft Research demo uses 5 cameras), or maybe in very particular cases (like a pure pan without object motion as a source).
However often you can do a good job by cutting the image into regions with 3-4 roto masks, setting the alpha to eg 0.25, 0.5, 0.75 and 1.0 and then fiddling with that in SmoothKit ZBlur as your starting Z map base.