Wednesday, October 08, 2008

Gigapixel depth of field

Continuing to consider the future of the DSLR: how will future cameras cope with the increasing diffraction limit?

The diffraction limit is a softening of focus that occurs at small aperture sizes due to the physics of waves. The actual point at which a camera is diffraction limited depends on the pixel pitch on the sensor. According to one calculator, a 12MP DX format camera is diffraction limited below f/5.6, while a 12MP FX format sensor is diffraction limited below f/8.

These are not particularly small aperture sizes. In fact my main DX lens is f/5.6 meaning that I have to keep it wide open to completely avoid diffraction effects. Fortunately the effects are gradual and only become noticeable/uncorrectable at f/11 or so and 100% magnification. But it puts a hard limit on pixel resolution for that particular lens at 25MP or so.

Going to larger pixel resolution will require larger apertures which will in turn limit depth of field. That's fine for portraits where shallow depth of field is usually the objective. But for landscapes and architecture, deep depth of field is more likely the desired effect. What is the use of 80MP if you have to use f/2.8? Wide angle lenses help of course, but sometimes the desired effect requires a narrow field of view.

One answer to the problem is to increase the sensor area of course. And that will be one of the reasons that Nikon and others have returned to the full frame sensor format. But that only postpones the problem.

What I expect will be the eventual solution is to adopt a technique used for may years in macro photography: combining a sequence of pictures taken at different focus distances. Today that is a technique that requires the full version of Photoshop or similar. But there is no reason that it could not be applied in the camera.

A secondary benefit that might be made use of is that a potential byproduct of the process is a 3D image map. Its not quite stereo vision (the picture is from a single point of view and does not contain the same information as a two lens stereo vision camera provides, but its close enough to be faked in software.

No comments: