Showing posts with label Old World Wisconsin. Show all posts
Showing posts with label Old World Wisconsin. Show all posts

Sunday, February 24, 2013

Rescuing REALLY IMPORTANT snapshots from overexposure

Last summer Charlotte took her mother, her grandmother and me on a time travel adventure to Old World Wisconsin. She'd been there before on a school outing and promised to show us the really good places. I could even take my camera if I promised to document our adventures.

One of the good places was the blacksmith's shop where, as the only kid, she got to do all the fun stuff like working the air pump to heat up the forge fire so the blacksmith could bend an iron bar into a hook. She  even got to buy S hooks like she help make to hang up her REALLY IMPORTANT stuff in her bedroom As promised I documented her every move with my D7000  set at ISO2000.

Then it was across the road to the German farmhouse where it was washing day. She got to help here too, doing everything from pounding a shirt in the rinsing tub to twisting one end as she and the old timer haus frau wrung it out. Again, as promised, I documented every step. Then, horror of horrors, I discovered I'd forgotten to reset the camera's ISO. All the pictures of this REALLY IMPORTANT part of the adventure were overexposed!!




Can they be rescued? Ask RawDigger. Like RawTherapee it's a free download found at  http://www.rawdigger.com/  It's easy to use and will tell you all the things about your raw photos that you never thought to ask.


RawTherapee's highlight reconstruction works by using the data in the less damaged RGBG channels to fix the data in the totally blown channels. In this RGB render mode, you are looking at a rough color image created without a true demosaic or white balance  You can also look at the composite mode which is very green or at the individual color channels.

The red splotches show the blown areas: in lighter red where there is data in at least one channel; in darker red where all data is blown out. RawDigger will also shows underexposed areas using blue splotches.

The program has another feature  worth the downloading time if you are ever faced with a directory of overexposed  images. With a CTRL + left or right key I quickly stepped through the overexposed NEF files to grade them from 'easy fix' to 'this will take serious work in GIMP or Photoshop.'

I graded this image as moderately difficult. While there are numerous bright red patches that will cause problems they are not in areas that would do major damage if they didn't fully clean up. So for this tutorial I'll concentrate on Charlotte's shoulder and an area in the closest tub.

Next question:  what highlight reconstruct method should I use?  The choices are:

Luminance Recovery



CIELab Blending


Blend


and Color Propagation


With this image Color Propagation is so obviously superior--the wood grain in the tub, the strap on the shoulder-- you might ask why are the other methods still on the menu. One problem is Color Propagation can produce a reddish color cast. Something that is fine for skin tones but not so fine for violet tinted skies. A second problem is it may not work as well with other camera brands.

It certain doesn't work with all camera profiles. This image would have been an 'easy fix' if it wasn't for the way Nikon compresses data to a supposedly lossless raw file. Except for a few speckles the blue channel is not blown. At the moment RT doesn't have a D60 input profile so I cleaned it up using the camera standard profile.



And this interesting monstrosity  is what happened when I switched profiles to the Nikon D60 Adobe Standard.


All profiles are not created equal.

Finally this is my rescue workflow.

Reduce the exposure. With this photo by about -2 EV.


Go to Highlight Reconstruction with Color Propagation and let RT do its magic.


Charlotte's face is too dark.  Go  to CIECAM02 and added a parametric Lightness curve


That's it, Not perfect but from what I saw with RawDigger what I expected  Once all color channels are blown the data is gone forever and not even RT can bring  back fine detail like the wood grain in the bucket.. But RT can reconstruct colors in areas like Charlotte's shoulder where there is no important detail,  Plus in the final rendering the transitions are not as sharp as they look in the view boxes.

Tell the truth. If you hadn't been following the tutorial  would you have noticed Charlotte's shoulder or the wood grain in the bucket in this final image.



I'd like to thank iliasG and DrSlony for the advice they gave me. While any mistakes here are mine, if they hadn't set me straight on a few key matters, lets just say things would have been embarrassing. Some of my assumptions were not, in political media jargon, 'true facts.'

edit- DrSlony, an RT developer, sent me his workflow;

Scribble I use the "Blend" highlight reconstruction mode by default. Most of my shots are landscapes, which means I usually have time to plan the shot and there are not unexpected over-exposures. The Blend mode makes the transitions from well-exposed to over-exposed smooth and it keeps the overexposed area white, which is usually what I want. Color Propagation is amazing at reconstructing overexposed skin, so whenever that's the case I use CP. CP is also the only method for which you can move the Highlight Recovery slider well past 100. When you do that with the other methods, the overexposed areas quickly get dirty grey or some other form of ugly, but with CP you can easily go over 200 or even 300 and get a great looking result. CP fails at large overexposed areas, and it's slower than the others.

Anyone else who wants to contribute his or hers experience with RT is welcome to join in.


You can find this build of RT at-
http://www.visualbakery.com/RawTherapee/Downloads.aspx
The online manual is at-
https://docs.google.com/document/d/1DHLb_6xNQsEInxiuU8pz1-sWNinnj09bpBUA4_Vl8w8/edit






Thursday, September 20, 2012

Teen with Fedora--A Second 'First Look'

In my original post on  RAW Therapee's new noise reduction routines I had excellent results with the image of Charlotte under Lord Shiva but far worse results with some other images. The noise reduction part still worked but thick black vertical lines appeared in the converted.images.

Turns out the noise reduction algorithm broke the image into segments called tiles. It worked its noise reduction magic on each tile individually but sometimes the tiles didn't rejoin smoothly. A tricky and hard to find bug, judging from the number of posts that appeared in the forum while they were fixing the problem.

That bug has been mostly swatted--noise reduction still doesn't work on JPGs-- but with code change set 135 we can now test and work with RAW files.  ( found at http://www.visualbakery.com/RawTherapee/Downloads.aspx  )



ISO 6400  f8.0  1/25 sec  EV -0.7

This snap was taken at Old World Wisconsin in the back of a barn during a demonstration of 19th century cow milking.  The lighting wasn't insanely difficult but as you can see from the camera settings I was pushing my D7000 to get this shot.

So how did RT do when I tried to improve the image.


YUP!  RT noise reduction works,

The noisy image to the left is without any noise reduction, just a +0.7 exposure correction to compensate for the -0.7 correction I did in the camera to move the shutter speed up to 1/25 of a second. The NR settings of the much cleaner image on the left are shown below.



I adjusted the Chrominance slider first. Use a 400 % box on something with little detail like the girl's cheek. Find a setting where all the color speckles disappear and then add a bit more.  This slider works on the A and B LAB channels that only contain low resolution color information so you are not blurring detail. With my D7000 at ISO 6400 about 2/3 of the noise was chromatic so that was a no fiddling around adjustment. Nice




For the other two sliders I picked a region with fine detail, the wood grain on the 2x4, and adjusted them to match the detail in my conversion to the detail in the jpg out of the camera. The Luminance slider controls the strength of the noise reduction. The Luminance Detail slider sets the threshold on  another algorithm that does its best to seperate real detail from random noise.  In this image Luminance 70 and Luminance Detail 50 worked well.




So where are we at now.  The left top line profile is from a RT conversion with no noise reduction.
The one below that is from the camera's jpg where I used the D7000's normal noise reduction setting.  It also is very close to the noise I measured when I used RT's default settings.

The top right profile is from my detail preserving image.  And finally the insanely low noise ISO 6400 profile below it is from when I went all out with a Luminance of 90 and a Luminance Detail of 6 to see how low noise I could go.

How bad was that image.  When I displayed it side by side with my tuned image I didn't have to pixel peek hard to see I had lost detail.  But if I had emailed it to Fedora Girl I'm sure she wouldn't have screamed "WHERE IS MY DETAIL" and immediately sent it into the bit bucket. We RT users are after the very best quality but in this world drowning in imagery, I'm afraid most viewer don't notice.

So great job RT folks and especially Emil who designed the algorithms and did the coding.