The interesting thing here is that everybody is at least partially right.
Inspired by your collective hints, I found an easy way to do better.
First of all, I had bracketed. The original image I posted indeed had the core of the red
tail lights blown. So was part of the moon. BUT... I had another image taken a couple of
minutes later at about 2.5 stops less exposure. Here, a typical tail light ranged from
about RBG = 255,160,80, to 192,79,61. But even so, the red tail lights still came up as
white when I "developed" the image to a JPG with a width of 1200 pixels in
Capture One.
But when I exported that same image to a 16-bit TIF, and then downsized it in
Picture Window Pro, things got much better. Sidways-viewed tail lights on the
freeway were red. Tail lights pointing at the camera on the entrance ramp were
orangy-pinkish, so they looked OK. And the moon is better in this picture.
<https://www.flickr.com/photos/24844563@N04/15876024034/in/photostream/lightbox/>
This is more like it, it looks like what I remember seeing.
The RAW development algorithm and the resizing algorithm are both "black
boxes." By using a different program to resize the image, I found a better black box
for this situation. What does a program actually do when it has to mash four pixels down
to one? Capture One (algorithm unknown) seems to emphasize the brightest pixel in the
bunch. Picture Window Pro defaults to a Bicubic algorithm, which seems to weigh the
darker pixels more, and I end up with something that looks like a bright tail light
instead of a white pinpoint. Me happy.
--Peter
On 2/10/2015 2:22 AM, Peter Klein and Moose engaged in frank and constructive
dialog:
We were treated to quite a show during our anniversary dinner.
<https://www.flickr.com/photos/24844563@N04/16304267069/>
I'd like some advice on a technical problem with this picture, and
in fact with all of the pictures I took of the cityfrom atop the
Needle. Red car tail lights turn white. It isn't overexposure
Well, yes it is.
, because it happens even on shots that I deliberately underexposed
drastically, where all the RGB values are under 255.
Are you sure of this, at the pixel level? If I throw a lasso around the
area of taillights on the screen sample, I get adecided stack of 255
values in all three channels at the top of the histogram of just that
small area. That's classicclipping, and will indeed turn red to
white.There is similar clipping of the blue channel of the blue lights I
checked, but no red and only a very little greenchannel clipping, so
they are probably reasonably accurate color.The difference between the
red and the blue highlights is the nature of the lights themselves. Most
taillights are stillincandescent bulbs behind red tinted plastic.
Although mostly red looking, they are wide spectrum, with a lot of
greenand blue. So when clipped a lot, all three channels clip, and 255
for all three channels makes a white pixel. The bluelights, OTOH, are
most probably 'neon' or LED, putting out narrow spectrum, almost pure,
blue light. So when they clip,it's only the blue channel and no harm
done.There are also clipped highlights in the relatively white office
lights, streetlights, etc. but you don't notice becausethey look
fine.Chuck is exactly correct. Very small areas of blown highlights are
just too small in number to show up in an overallhistogram. I often
select a small area or Select Highlights (a PS menu item) to check the
state of clipping.
In shots with any decent city detail, the red channel does hit 255,
but the green and blue are much lower. Loweringthe exposure in post,
or using highlight recovery has no effect.This doesn't happen with
larger areas of red light where the pixels have the same values as
the little tail lights.But taill lights turn white, as do other
points of colored light like the blue and green Seahawks colors that
stilladorn some construction cranes. The key seems to be that the
light sources are just a few pixels in diameter.
Indeed, that is the key, as Chuck and I have pointed out. Those little areas
are simply overexposed.
Maddeningly, on the onscreen preview, the tail lights appear red.
But when I "develop" the JPG, they turn white.Theyalso turn white if
I blow up the onscreen preview to 50% size or larger.
Size, sampling to screen resolution and your eyes at work. They are always
blown.
The camera is an Olympus E-M5 with 45/1.8 lens, and the RAW
developer is Capture One v. 7.1.2. Here's a screen clip ofthe whole
picture, followed by a section with and a string of car rears blown
up to 400% so you can see what's goingon with the pixels.
<http://gallery.leica-users.org/v/pklein/temp/SpaceNeedleMoonriseScrPreview.JPG.html>
I like the moon exposure better in this one.
<http://gallery.leica-users.org/v/pklein/temp/SpaceNeedleMoonriseCars400pct.JPG.html>
Advice, anyone? Is this just an inevitable result of the Bayer array, or is
there a way to fix it?
Nothing to do with a Bayer array. As Chuck outlined, there's no way to
fix it for this sort of scene. At least no simpleway.What I would try is
even greater underexposure of the whole scene, using a lower ISO, to
keep noise down as far aspossible. Bracket, and use the highest exposure
where the lights stay red at 100%. Then in post, raise the midtones,
anduse NR to control noise there. Separately whack the shadows with a
combination of masking them out of the increasedbrightness used on the
mids and really heavy NR.I've used this strategy successfully on other
subjects with similar problems. It is time consuming and I can't
imaginedoing it in an editor that doesn't have layers and layer masks.As
Chuck suggests, the other solution, without reshooting, is to replace
the offending pixels with correct colors. Youmay do it by retouching,
one by one. Or, in PS, select them by color and area and use replace or
Match Color. Either wayis likely to be tedious.
Red Moose
--
What if the Hokey Pokey *IS* what it's all about?
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|
