[popo]

Having read it, I think the understanding might be a but optimistic. My take on it is that DPRAW will store not only the light level from conventional raw, but also the phase information in some way. I don't know if they'll store the sub-values directly, or a derived level+phase value, since the phase has to be decoded already for the AF system to use it. Phase being some value showing if that point is in focus or not, and if not, how much in which direction. The roughly doubling of file size would be consistent with that.

 

Now, what could you do with that data is another matter. Adding extra blur according to the defocus amount could more accurately simulate bigger apertures than without that information, but I'm less convinced in refocusing. There are ways to improve images if you know the nature of their spread (e.g. by defocus), but in my experience they aren't good enough for high grade imaging, and are particularly sensitive to noise.

 

If I'm not mistaken, previous dual pixel implementations have been in a single direction have they not? I would think they would need a matrix of at least two directions for improved performance.