Post Processing Challenge with a Curve

This is another underexposed image. It was underexposed on purpose so as to not blow out the highlights in the shirt. The resulting histogram is actually pretty good. But the subject is way underexposed. The pic was actually shot in very bright sunlight ISO 100, f/4.0, 1/500 sec.

The challenge is to improve the image using only curves and sharpening. The image was converted to sRGB.

HTTP response: 500

Oh man.... homework?!

I'll try this one (when I get home from work) ... it's a lot like what I get when shooting the Egrets.

OK, I think I went in circles a bit. I'm not so good with color casts I think, and since Maderito said "curves only" that really threw me for a loop.


I didn't like some of the red, so I wanted to bring it down a bit. Curves, Red channel, 71, 65 ; 131, 124 ; 255, 248

That brought out some greens of course, so Curves, Green channel, 255, 248

Now I went to lift out the darkness: Curves, RGB, 47, 84 ; 126, 148 ; 208, 205

Well of course that brought out the blues, so Curves, Blue channel, 15, 0

Finally, my sharpening was USM at Amount 134%, Radius 0.3, Threshold 0

And save for web. Having such a small JPG as a starting point really stinks! Way too many anomalies for my eyes. So I went with Quality 80, but also added Blur 0.28 to reduce some blockiness, but this also destroyed a bit of my sharpening.

The final result - I didn't dare go much further as things started looking bad:

I used three curve corrections, all on adjustment layers.

1. Color cast removal
The cast is not obvious. Place a sample point in a shadow of the white shirt. The one I selected was R=114 G=123, B=129
The average luminance is 122. So we'll neutralize that point to Gray=122 and thus remove the small blue cast.
This is important since neutral midtones will stay gray as we make other bigger changes.

- Make a curves adjustment layer, name it "Color Correction"
- Set the following points near the midpoint of the individual R/G/GB channel curves (you can use your own image sample point or enter these values from the one I used):
Red channel curve: 114, 122
Green channel curve: 123, 122
Blue channel curve: 129, 122

We could have accomplished the same thing by using the Set Gray Point eyedroper in the curves dialog.

2. Gamma correction
The image is way too dark. You lighten by increasing gamma. But you want to lighten without blowing out the whites of the shirt. Thus the gamma adjustment should affect the shadows more than the highlights. In levels, you increase gamma by moving the middle slider to the left. In curves, you bend the curve upward and to the left.

- Create the second curves adjustment layer and name it "Gamma Correction".
- Set 2 points in the main RGB curve, one to bring up the overall brightness weighted towards the shadow areas, and the second to reduce the correction in the highlights so as to maintain detail in the whites of the shirt.
RGB curve point #1: 34, 75
RGB curve point #2: 179, 122

I then sharpened the image but it still seemed a bit dark, so I did one final adjustment to brightness.

3. Final brightness correction
- Make a final curves adjustment layer and call it "Lighten".
- Set the following point near the midpoint of the RGB channel:
RGB curve: 111, 133

Almost done.

When bringing out detail in a severely underexposed image, you get a lot of noise, especially in the darker portions of the image. Here's a simple, well-described method to reduce this noise.
4. Noise reduction
- Duplicate the original background layer.
- Blur the background copy with a Gaussian Blur, radius=5.0.
- Set the blending mode to color.
- Merge the two image layers.
- Sometimes you have to increase saturation in the final result, but I didn't think it was necessary here.

HTTP response: 500

maderito wrote:
This is another underexposed image. It was underexposed on purpose so as to not blow out the highlights in the shirt. The resulting histogram is actually pretty good. But the subject is way underexposed. The pic was actually shot in very bright sunlight ISO 100, f/4.0, 1/500 sec.

The challenge is to improve the image using only curves and sharpening. The image was converted to sRGB.
http://display.lifepic​s.com …ptq%0C8289%3BB%​0F4%3A3846

In bright sunlight? You? Or the performer? Or both? I have a feeling there's going to be a bit of color noise in this image when it's "fixed" but I'm going to give it a shot anyway when I get home tonight.

This is fun to see how differents are the results.
Here is mine using uleadphotoimpact and some
neatimage filtering.
Jose

HTTP response: 404 | MIME changed to 'text/html' | Byte size: ZERO

I duplicated the image twice in screen mode. Masked the pants once and the shirt twice. I then adjusted the bright/contrast slightly untill I was satisfied.

Here's my attempt:

1. Create a duplicate layer of background.

2. Make a selection out of the performer - do not use feathering when selecting.
- After making selection, "Select | Modify | Expand" 1 pixel.
- "Select | Feather" 3 pixels.

http://www.tanseikai.c​om …ges/others/imgd​isp-B1.jpg


3. Work on the main performer:
- Create new Curves adjustment layer:
("Layer | New Adjustment layer | Curves...").

- Set points as indicated:

http://www.tanseikai.c​om …ges/others/imgd​isp-B2.jpg


- Use Gray eyedropper to remove bluish color cast.

4. Now work on the background. Switch the selection to the background:
- "Select | Reselect"

- "Select | Inverse"

5. Create another Curves adjustment layer and set points as indicated:

http://www.tanseikai.c​om …ges/others/imgd​isp_B3.jpg


6. Select the Background copy layer from the Layers palette.

7. Sharpen as desired. I used USM:
Amount= 100%
Radius= 1 pxl
Threshold= 3

8. Flatten layers and save.

http://www.tanseikai.c​om …ages/others/img​disp_B.jpg


And, BTW, here's a better rendition (I think) using 2 steps:

1. Use levels to remove color cast.
2. User Shadow/Highlight to finish "fixing" the image.

You can't make it any simpler than that!

http://www.tanseikai.c​om …ages/others/img​disp_C.jpg

Here's my attempt:

Byte size: ZERO | Content warning: NOT AN IMAGE

Oh, I forgot to sharpen the iumage.

Byte size: ZERO | Content warning: NOT AN IMAGE

A difficult image and an interesting range of approaches to improving it. It really was shot in sunny daylight but underexposed by 3 stops.

I will post later today or tomorrow one additional post processing result for this image. Since I have the original RAW file, I can cheat a bit. I will use a "hybrid conversion" technique described here: http://www.luminous-landscape.com …s/hybrid-conversion.shtml . It combines linear and standard RAW processing. I'll do this using only curves for tone and contrast adjustments. Perhaps I'll use the a sharpening method from the gmitchel toobox.

I may post the result on a followup thread since this one is overloading my DSL connection.

Cheers until later...

maderito wrote:
I will post later today or tomorrow one additional post processing result for this image. Since I have the original RAW file, I can cheat a bit. I will use a "hybrid conversion" technique described here: http://www.luminous-landscape.com …s/hybrid-conversion.shtml . It combines linear and standard RAW processing.

The basic idea of hybrid conversion is to capture the full dynamic range of the CMOS RAW image data by using RAW linear processing which does not involve clipping of any channel data. The resulting very dark image is then gamma corrected, converted to B+W, edited for tone and contrast, sharpened, and finally converted to grayscale. The image now contains corrected tonal values minus color.

The color information in the image is then developed through routine (i.e. non-linear) RAW processing, with emphasis on extracting color information (hue and saturation) and not worrying about brightness/contrast. This second image is converted to LAB mode which has one luminosity and two color channels. The luminosity channel image is replaced with the B+W grayscale image (via "Apply Image"), thus merging the tonality of first image conversion and the color of the second image. The final result still needed further editing and adjustments. The result was pretty good.

HTTP response: 500

Pretty good?

It's hard to believe that the above image was derived from the original image posted.

I'm hooked.

I just tried using the hybrid conversion method on a few of my backlit images that were on the underexposed site and man oh man, what a difference. The hybrid conversion method really does do wonders to "hopelessly" underexposed images.

Thanks, Woody, for bringing this to our attention. This is definitely something I'm going to be using a lot of, that's for sure.