Underexposed Image #1

RGB Original

Lab Correction

Meet a typical underexposed image, a diamond in the rough that, in spite of its dark appearance, still has plenty of image data compared to a similarly overexposed image. The Lab corrected image on the right (ignoring the jpeg artifacts accompanying this image) has good overall brightness, color differentiation between the clothing and blackboard, and the flesh tones are square in the middle of what we'd expect. Read on to see how it was done.

Lab is the image space extraordinaire for underexposed images. With any image that will require large brightness moves, Lab mode is the first color to go looking. Later, we'll give RGB a shot, and see what the typical problem is with that color space, and get a surprising result, indeed, when that problem is addressed.

First, the Lightness channel. There is no neutral highlight, so the overall image was lightened by sliding the white end of the Lightness curve inward to light up the specular reflections on the eyeglasses. I lifted the middle of the Lightness curve a bit to increase detail in the clothing and brighten the image overall.

You may download the Lab curve file used to correct this image.

Next, juice up the color with the a and b channels. To get rid of the image's greenish cast I drew on my many hours of observation in the classroom, took a chance, and dropped a neutral point on the blackboard in the background.

Note: Had this been a green backboard, the entire image would have taken on a magenta cast. As luck would have it, no magenta cast appeared, and there was the added jackpot of the other colors in the image improving as well.

Underexposed images are generally undercolored images as well. To kick in some color, I rotated the a and b curves around the neutral to get full flesh tones. The upper right of the a and b curves, which control the flesh tones, were then pinned in place, allowing the lower left end points of the a and b curves to be rotated to punch up the color in the clothing.

RGB is the silk route of color spaces, All color images, one way or another, must pass through RGB. This does not automatically make RGB a good color space for color correcting, and this image illustrates the main reason for this.

Here are two automatic corrections from Photoshop, Auto Levels, and Auto Contrast. Neither is very good. Auto Contrast gives a very serviceable result. If you are on deadline, and have just a couple of seconds for each image, you could do worse than applying Auto Levels, and Auto Contrast for each image, and picking the better of the two.

Auto Levels

Auto Contrast

Can we beat Auto Contrast?. It's time for some curving, so let's do what most of us do the first time we run into an underexposed image: raise the RGB master curve in an attempt to kick this image into gear. The result, RGB #1is not really all that bad, and in fact it looks a lot like Auto Contrast.

RGB #1. Using curves as if it were the Levels command, I dragged the white corner inward until I tuned in a pretty good image.

RGB #2. If we keep dragging to the left,the image gets light enough, and we get a color correction train wreck.

What's happening with the two images above? We are using the master curve to promote red, green, and blue equally. In RGB #1, the red channel is just on the verge of blowing out. Not bad. The image is still too dark overall. It's still a little dark, so ...

If we then push our luck, as in RGB #2, the image brightness is better. Look what happened to the colors in the face! The green channel starts to top out, and we get broad areas of yellow in the flesh tones, and the entire image is threatening to go green on us. This is RGB's Achilles hell - er heel. Brightness and color are mixed together. There appears to be no way out of this dilemma. This experience, by the way, is the reason so many of us run screaming at the thought of ever using curves again, which is too bad because there are a lot of images out there that could benefit..

But we know that a good image is possible, and even easy, in Lab mode. So it stands to reason that there is more that we can do in RGB mode? Well, the first thing is to recognize that the RGB or master channel always causes color casts, and should almost never be used for large curve moves, as we have done here.

So let's ignore the master channel for now and adjust the curves separately. Wow - the result is not shabby at all! The Lab corrected version is shown once again for comparison.

RGB adjustment using the curves on the right	The curves above took a very minimal amount of time, being basically a variation on Auto Levels, with two changes, described below. Load the RGB curves here.
Lab adjustment, shown here for comparison

The above curves were created in three steps. Before doing anything else, and knowing that a neutral blackboard was the key to the successful Lab image, I dropped a neutral point on the blackboard, resulting in the middle curve points that you see.

I then dragged the upper right curve points to create a straight line with the neutral points. This resulted in a significantly brighter image, with very red flesh tones. The upper right two points of the red curve reduce the amount of red in the face, while retaining most of the benefit of the area surrounding the red curve's neutral point.

Finally, I dropped the middle of the RGB master curve a bit to darken the image overall, and add contrast to the face.

The result is much better differentiation in the clothing color, good detail in the face, albeit with a bit more a suntanned look than is probably appropriate. I still prefer the Lab version because of the better skin tone and contrast, and the RGB version has a lot to recommend it. Do you have a favorite?

Underexposed Image #2

First, as with most underexposed images, start in Lab, and set highlight, shadow, and neutral. Luckily, this image has all three of these. Angels - at least the ones we pin to our trees - are known to be pure white. The woman's sweater is probably black or at least very dark, and the wall behind them is probably neutral. Let's go with these assumptions, and see if we like the result:

Lab #1 Here's the result of following the Color Wizard's procedure of mechanically setting shadow, hightlight, and neutral. The resulting image is better, and still a little dark. The image also exhibits the flat colors that are typical of underexposed images. The leaves of the plant are almost gray. Note that the poinsetta has more color due to being closer to the flash. --- Load the curve for this image here.

Lab has some tricks to deal with both of the issues that are bothering the above image. The remaining overall darkness is easily dealt with simply by dropping the Lightness curve slightly (in this example we have black on the right side of the curve, instead of the usual left).

The most interesting thing we do is with the color capabilities of Lab. The a and b channels of Lab represent a total of four curves, one each for magenta, green, yellow, and blue. We work these puppies for all they are worth as follows:

Load the curve file for Lab #2

Lab #2 Load the curve file for Lab#2 and look at it in Photoshop. The Lightness curve move is a very standard one, consisting of dropping the middle of the curve just a bit to lighten the image overall, and add some steepness to the darker half of the curve. The action in the a and b channels takes a little more explaining. First, the flesh tones are stabilized by adding anchor points near the middle of the a and b channels. Blue color is pumped into the woman's shirt by steepening the upper right corner of the b channel. The red of the flowers is increased by steepening the extreme lower left corner of the a channel. Lastly, locking the magenta corner , and steeping the green corner of the a channel has one and only one effect - it adds green to the foliage. Levelling out the yellow corner of the b channel likewise adds punch to the green.

The banding of the wall may be addressed by adding gaussian noise in Photoshop or Elements. Other problems remain - for example the flesh tones could probably be further improved. Nevertheless, the overall result breathes life into what must have been a happy moment for all concerned.

After walking away from this image for a while, and coming back, I realize that it was probably an error to assume the wall was neutral (in fact it was probably green), and this is why the faces are the wrong color. No matter. Let's tackle the flesh tones in another tutorial, "Pinning Flesh Tones".