D90 - what is possible in post production...

i didn't like the sharpening, it looks like an image put through a sharpening filter and not a sharp image, but it's nice to see that people are experimenting. i also think there's a lot you can do. i'm currently experimenting with warping, statistical and fractal approaches to sharpening, it's quite promising. i'm also thinking of, in the same spirit as the stairstep removal, taking advantage of the fact that the debayer algorithm seems pretty lousy and maybe reconstructing the raw file and redoing it, only better. if you look at the channels individually you notice that they are more different than they are supposed to if debayering was done correctly, causing color aliasing at all lines. in other words there's more luma information in the image but we don't see it since it's hidden in chroma errors.

keep it up.

/matt
 
mattsand said:
...i'm currently experimenting with warping, statistical and fractal approaches to sharpening, it's quite promising. i'm also thinking of ...taking advantage of the fact that the debayer algorithm seems pretty lousy and maybe reconstructing the raw file and redoing it, only better...
Click to expand...

Please, tell us more. I don´t get exactly what you mean. Yes, the sharpening I have done, is a bit extreme. But it should show what is possible. This music video seems to be very clean:
http://vimeo.com/2678480
What do you think they have done in post to get it out so well?
 
Just like Matt just said, it does look like a sharpening filter was thrown on there. I like the before more then the after. I wish Nikon did better implementation of the movie mode as does all of us here, but we gotta work around this wonderful camera and its abilities.

And..Wow that music video is clean...Well put it this way, when your shooting during the day, and you have a good lighting setup, the D90 will be nice to you and give you a real low ISO.

I think I'd even go as far as saying that it's like shooting on film? It looks horrific in low-light, or just un-recordable. If you're shooting indoors, you need controlled lighting BIG time, as intense as possible, because D90 will kick up noise if it feels theres not enough light around.
 
I disagree completely, the low light performance is excellent (what om earth are you comparing with?) and while lighting is always a good idea to get a nicer image regardless of iso issues you definitely don't need a lot of watts. Just try to not overexpose, the auto exposure tends to, and you'll be fine with a couple of 40w bulbs. /matt
 
ok, i'm gonna tell you more now: i believe the camera uses a very crude debayer algorithm that simply grabs all the red, green, and blue pixels to their respective buffers and linearly or nearest neighbor interpolates them to full size. if this is true it's easy to reconstruct the raw bayer pattern, by simply doing a nearest neighbor downsample, and debayer using a better algorithm that's sharper. the "only" complication is that the data has been compressed, but it might still work.

here's how it works, the camera seems to use one of these, neither is perfect but both create reversible results so we can go back and redo it:

http://winfij.homeip.net/maximdl/bilineardebayer.html

/matt
 
do you have used it?
how to use it?it is a plug-in but i don't understand in wich program i have to use it.virtualdub??
do you have images?
thanks
ivan
 
...to begin with i'm gonna try to use bicubic interpolation of the channels to see if there's an improvement, and if there is i'll move on to an algorithm that uses info from the other channels to make educated guesses about the current one. while the color level of a pixel varies with the color of the subject, the gradient doesn't as much. if there's an edge in the green channel we can be fairly certain there's supposed to be one in the blue channel too in that pixel position, even though the camera didn't record it and even though the blue channel itself interpolated might suggest a smoother gradient. this is probably way over the head of most people, but i thought it could be interesting and maybe somebody wants to add something? :-)

anybody has an idea of which comes first in the d90 video image, a green/red or green/blue line and whether it starts with a green pixel or not?

problem left to figure out: does it debayer before or after the lousy stairstepping scaling? i'm guessing before since the stairsteps are easily the sharpest detail in any image and i don't think a lousy debayer interpolator could produce that sharp an edge. thus, a proper rescaler should be applied before the re-debayer. fortunately i already got one. :-)

/matt
 
shiny4 said:
do you have used it?
Click to expand...
that was to illustrate how to not do it, how i believe the camera is currently doing it. :-) no, there's no plugin or software to use, it's in my head for now and it will probably become either an fcp plugin and/or a unix shell script in the end. stay tuned.

/matt
 
mattsand said:
.. i believe the camera uses a very crude debayer algorithm that simply grabs all the red, green, and blue pixels to their respective buffers and linearly or nearest neighbor interpolates them to full size. if this is true it's easy to reconstruct the raw bayer pattern, by simply doing a nearest neighbor downsample, and debayer using a better algorithm that's sharper...
Click to expand...

Oh my god, that´s way too much over my head. My technical knowledge is very limited. But I wish you good luck mattias because we all will probably profit from the outcome if you can manage it... :-)
 
haha, check out that link though. the solution is a bit difficult but everyone using digital video in post production should know the concept. the sensor captures a black and white image with a color filter applied to each pixel, then you construct an rgb image from it using interpolation. there are simply worse and better ways to do it (possibly some great too).

/matt
 
You must log in or register to reply here.
Back
Top