|
Forum: Canon, Nikon, Sony, and Other Camera Brands
07-26-2012, 01:33 AM
|
| |
You have no idea what are talking about.
Facts:
1) CDAF doesn't have any concept of focus points, it works until some contrast measure will achieve the maximum (today they uses shortcuts as well in order to minimize focus hunting and computations). Due to its nature CDAF is the most accurate method. It also benefits from some sort of scene and pattern recognition.
2) PDAF uses fixed placed sensors, so we only have "focus points" with this method. With this approach we can determine we either have back or front focus at the moment as well as defocus estimation which can be translated into the approximate defocus distance (some sort of indication). You see this method can't be particularly accurate, because it greatly depends on lens and sensor alignment. It also has no idea about the scene (although advanced DSLRs use their metering sensors for scene estimation, but this is quite an expensive approach, both cost and power consumption wise). This method is faster to nail approximate focus since it excludes hunting at all.
The conclusion is Nikon 1 focus points won't bring anything to get more accurate focus. Instead, they help to nail it faster. Hunting is still in final stage when CDAF in action, but the amount of this is quite negligible.
PS There are some new researches that tries to perform approximate focus estimation (i.e. to get focus direction and approximate defocus distance). This problem cannot be resolved with high precision due the nature of model they uses: I'=A*I, where the I is the properly focused image we want to get, I' is the image we have at the moment, the A is so called "spatial convolution" operator, which is also called "S-transform" and represents lens caused blur. Unfortunatelly, the A cannot be reversed otherwise the problem will be quite trivial (if the A would be reversible we will be able to recover missed focus as well during PP, in this case focusing algorithm would act like virtual CDAF). They are trying to make defocus assumptions based on image recognition technics and S-transform properties. They have quite succeeded already, but this isn't production ready yet, since it needs high amount of computations.
|
