"Thomas de Groot" <tDOTdegroot@interDOTnlANOTHERDOTnet> wrote:
> That overlaying of images intrigues me. It is merging of two or more images
> in a paint program isn't it? What is used in that case? Averaging?

I geuss I'd do it with *POV* - because (a) it allows for best-quality input,
using HDR file format, and (b) I know for sure that it will use the maximum
possible precision for computation - using an orthographic scene process; and
yes, that would be plain averaging. Which is exactly what each instance of
mcpov does anyways for all of its iterations (although that's done even at
float precision; but when parallelizing mcpov runs, we're probably talking
about just 4 or 8 shots that finally need to be mixed).

One thing to pay attention to would be to take runtime differences into account
appropriately. For example, if three threads had been running for 4 hours each,
and another render was started 2 hours later and consequently ran only 2 hours,
I'd give that 2-hour shot a lower weight.


If using Photoshop instead, there's two possibilities to go; both involve just
plain normal layer combination, with some transparency to the layers:

(a) process the shots in pairs, giving the lower layer full opacity and the
higher one 50%. If, say, you had shots A, B, C, D, E, F, G and H, you'd first



(b) stack all the shots; give the lowest layer 100% opacity, the next higher one
50%, the next higher one 33%, then 25%, 20%, 17%, and so on. Merge to a single
shot.

I have no idea which of these approaches would leave you with the least loss. I
guess (a), but it depends on how Photoshop does things internally. If for
example it stays in the 8 bit integer domain for its computations, (b) is
obviously crap; but if it uses floating-point math (or at least 16-bit
arithmetics) during any "merge layers" operation, but converts back to 8-bit
after such an operation, you're definitely better off with (b).

In any case, Photoshop can never do as good as POV.


Additionally, with POV you can stay in the linear domain (speaking of gamma
here) throughout the whole process before outputting the final results; I don't
know whether Photoshop takes gamma issues into account when merging images. This
will probably not be an issue when images are low-noise already, but if you
still have some deal of graininess it may make a difference.

Note however that when working with linear output from the actual renders, you
*do* want to use HDR, otherwise you'll lose a lot of detail in dark parts of
the shot.

Post a reply to this message