A recent discussion of ambient light underwater photography in the Wetpixel forum pages presented a wealth of information on the use of filters to correct color balance problems in photographs taken underwater without strobes.

But filters cannot fix color problems when shooting in ambient light at depths where red and yellow wavelengths of light are completely absent, not merely attenuated. Alternatively, the subject may be too large for conventionally mounted strobes, or the photographer working with strobes at depth may be unable to approach the subject close enough to accommodate the limited illumination range of strobe lights underwater. In either case, the result is that the subject is illuminated solely by ambient light in which red and yellow wavelengths may be deficient or totally absent.

The image recorded electronically or on film often looks even worse than our subjective recollection of the actual scene. This is because the human optical system is very good at adjusting to color casts that dominate the visual environment. When our eyes are flooded with a light of a certain color, the visual system compensates by becoming less sensitive to that wavelength.

The camera simply records what it sees and reproduces the wavelengths of light that enter the lens as faithfully as the design of the film emulsion or CCD sensor allows. When the printed image is viewed under normal illumination or an electronic image is displayed on a calibrated monitor, however, our visual system no longer compensates for the prevailing color cast, so the image exhibits a much heavier blue and/or cyan color cast than expected.

Examples can be found in the pages of almost any dive magazine. Images of cetaceans, sharks and many large pelagics taken in open water are classic cases. For really bad color casts, flip through any shark ID book. But you don't necessarily need to throw away that once-in-a-life time shot. Adobe Photoshop contains an array of powerful tools for correcting color casts. This tutorial introduces one of the most useful techniques for dealing with extreme problems, known as channel mixing, or “plate blending”.

Mola in the Blue

This shot of an oceanic sunfish ( Mola mola ) is a case in point. When we approached, the mola initially headed out toward open water and then turned back toward the reef, giving me a brief opportunity for a head-on shot. Unfortunately, however, the mola was about 10 meters away when it turned, beyond maximum strobe range.

The image was taken with the Nikon AF-S 12-24 zoom on the Fuji S2 Pro in a Subal housing, FE2 dome, +2 diopter, aperture f4.8 and shutter speed 1/45.

This is what the shot looked like after I had done as much as I could with the various options available in the Fuji EX 2.0 Raw-t0-TIFF conversion program.

I used a steep custom tone curve in the Fuji EX program and placed the gray picker on a spot on the front of the mola to set the color temperature. I also added +1/6 stop to the exposure. I tried to dial down the blue and kick up the reds as far as I could, selecting “fine tuning” color settings of R:1.14, G:0.99, and B:0.88, but as you can see, these gestures were largely futile.

(Note: The raw converter in the new Adobe Photoshop CS offers much more powerful tools for adjusting white point, color balance and exposure than the Fuji EX converter. It will do a better job, but it cannot generate an optimally balanced image from a difficult raw file such as this one.)

This was a cloudy, overcast morning and the subject was beyond strobe range, and the depth was about 28 m. A heavy blue cast pervades the entire image, rendering the large near-white area on the bottom part of the front of the mola's body a sickly cyan.

I was aware of the strong blue cast while looking at the mola in the water, but what the colors my blue-adapted eyes saw were much closer to neutral, ranging from near-black through brownish greys to near-white. But all the camera's CCD sensor could see were blues and cyans.

Finding a curve to match a mola

One doesn't run into Mola mola every day, and I may never get another chance to make a head-on shoot like this again. But the usual Photoshop techniques don't work well with extreme problems such as this.

Image > Adjust > Curves can a more powerful tool for adjusting color casts. Changing the combined RGB curve into a more steeply sloped S-shape does force more contrast into the image, but now the light-colored highlight areas have turned a glowing turquoise cyan.

If you're familiar with LAB color space, you'll know that converting to LAB before applying curves is often the best way to deal with extreme color casts. Shifting the A and B curves in LAB upward is a tried-and-true method to reduce an excessive cyan cast and boost reds and yellows. The result (C) is better than in RGB, but it also introduces an unnatural purplish magenta cast that pervades the entire image and poisons the background water.

Using the color balance tool reveals something interesting: moving the top “cyan-to-red” slider all the way to the right – which one might expect would kick up the red values – has no apparent effect on the image except to darken it.

Using Image > Adjust – Hue/Saturation to tone down the cyans and enhance reds produces similar results to using Curves in LAB.

Desaturating all the way to monochrome grey, followed by a quick correction of contrast using Adjust > Curves is a time-honored technique to salvage an image with excessive color casts such as this one, if you're willing to give up color and accept a black-and-white image.

Image > Adjust > Aut0Levels is often dismissed as a “beginner's tool” and is rarely useful in dealing with underwater images. This one is no exception.

But here's a surprise. Using Image > Adjust > Autolevels after the Hue/Saturation adjustment (see D above) produces startlingly good results:

When this tutorial was originally written, I was using Photoshop 6. I discovered this surprising effect while re-working the images using the new Photoshop CS. I didn't get a result anything like this when I tried these two steps together with the older version of Photoshop. It may be that I used different values when I adjusted Hue/Saturation before, but I suspect one or both of these tools may have been improved in the new version of Photoshop.

In any event, this combination clearly works under CS, and may be as good or better than the plate blending technique this tutorial is supposed to be about.

Reconstructing a new red channel

Examining the individual Red, Green and Blue channels of the original RAW-to-TIFF conversion provides important insights into the nature of the problem. The red channel in the original image isn't just bad – it's non-existent, a featureless black plate. Conventional curve boosting cannot correct the red values of this image, because there aren't any - just sprinkles of noise.

Red

Green

Blue

Extreme situations call for extreme measures. If we can't fix a non-existent red channel, we can still build a new one. In this case, moreover, there is no data in the red channel to protect, so we can throw it away without qualms. Our starting point is that we can assume the colors of the mola should be close to neutral (white, grey or black). RGB is a balanced color space, which means neutral colors are composed of roughly equal amounts of red, blue and green. The RGB values for the deepest black your monitor can display is {0,0,0} while a medium gray might be {130,130,130} and the brightest white is {255,255,255}.

CMYK, on the other hand, is not a balanced color space, so neutral colors require different amounts of cyan, magenta, and yellow, plus additional black (K) ink determined by the gray component replacement (GCR) value.)

Since this is an RBG image and the main subject is basically neutral, the values of the pixels representing the mola in our RGB image should roughly be similar in each channel in order to produce the near-neutral greys, whites and blacks we need. In the original conversion, the brightest part of the mola - which we assume should be close to white - is actually {0, 206, 235} while the darkest shadow on the mola is {0, 60, 85). What we want are red values for each pixel that are closer to the green and blue values. We can get these values by borrowing data from the green and blue channels to make a new red plate.

As is often the case, the weak green channel has better contrast and less noise than the blue. Blue channels are particularly suspect if the image has ever been saved in a JPEG format because the JPEG compression algorithm does a lot of its work by throwing away information in the blue channel.

There are two ways to carry out plate blending operations in Photoshop. One is Channel Mixer and the other is Apply > Image. The Channel Mixer is easier to use, but Apply > Image allows us to flexibly select among a variety of blending modes. When there is important, usable information in a damaged channel, these modes make it possible to preserve that data while enhancing or replacing missing or deficient color values with data from other channels, so it's worthwhile learning to use Apply > Image.

Quick and Dirty Guide to Plate Blending (or Channel Mixing)

First select the red channel on the channel menu in Photoshop and then click on the box to the left of the composite RGB channel so that you can view the effects of alternative blends in full color. Now invoke Image > Apply Image and change the source to the green channel. Since there is no data in the red channel to damage, the blending mode can be “ Normal ”, but “Lighten” or another mode may be a better choice if there were any data in the red channel worth protecting. Now adjust the opacity level to fine-tune the correction. For this image, I first applied a 75% blend from the green channel, followed by a 25% blend from the blue channel in order to bring out the slope of the reef looming in the mist in the lower right part of the frame. (The reef is only visible in the blue channel.)

This may not look great at first glance, but the difference is now that we have a usable red channel to work with. When we convert to LAB and apply Image > Adjust > Curves to increase contrast and tone down the magenta in the water, the result is very different than when we tried it before.

Get to know LAB

Now may be a good time to briefly discuss the LAB color space. LAB may seem confusing at first, but it is usually easier to correct color casts in LAB than RGB.

Instead of Red, Green and Blue, the LAB channels are L, A and B. The first LAB channel, L (Luminosity or Lightness) carries all the information about the darkness or lightness of each pixel – it's basically a black-and-white version of the image. All color information is carried in the other two channels. The A channel values represents the relative redness or greenness of each pixel. Shifting the curve upwards builds up reds and weakens greens. The B channel does the same for yellow versus blue. Altering the slope of these curves changes color contrast, while adjusting parts of the curve selectively changes different ranges of colors.

You can even invert the curves – changing reds to greens and yellow to blues, and vice versa.

It's obvious that every pixel in the new blended image (L) is more blue than yellow. However, the magnitudes of the B values of the mola pixels in the new image are less than the B values of the water pixels. This means we can adjust the B curve so as to reduce the “blueness” of the mola pixels while leaving the background water relatively unchanged. There is a slight difference between the mola pixels and the water pixels in the A channel as well, so we can selectively increase the redness of the mola pixels while bleeding reds and magentas out of the water by raising the left side of the A curve while dropping the right.

Lab L

Lab A

Lab B

Steepening the L curve in LAB is another valuable tool, because it increases overall contrast without changing colors, which can be difficult in RGB or CMYK.

LAB has a larger gamut than RGB or CMYK, which is why Photoshop uses LAB as its “home” color space. (When you convert an RGB image to CMYK, Photoshop actually converts it from RGB to LAB and then from LAB to CMYK.) That's also why it's important when working in LAB to make sure you have not generated colors that will be out-of-gamut in the final destination color space.

Last but not least, LAB is the best color space for sharpening. Select the L channel on the channel menu first. This avoids the risk of generating contrast halos or shadows in an unwanted complementary color, which sometimes happens when sharpening in RGB color space.

Final Adjustments

We're not quite finished yet. There is some snow and considerable noise in the background water. The gross adjustments we just made in LAB also need to be fine-tuned to further reduce the magenta tones in the background water, which might be better done back in RGB color space. And we need to do a bit of sharpening, using the High Pass and Unsharp Mask filters.

Next was to convert back to RGB and remove the remaining excess magenta in the background water color by dropping the RED curve in Image > Adjust > Curves. I used the clone tool to remove the largest bits of snow and sensor dust marks. Then I did a selection to mask out the mola, and applied Gaussian blur to smooth out the remaining noise in the water.

Conclusions:

Here is the original we started from, the final plate blended image, and a final image generated via the alternative procedure outlined for Image (G) above.

Some may consider plate blending an excessive modification of the “original” image taken by the camera. For me, however, the concept of an “original” image itself is problematic. We shape what the camera sees whenever we use a wide-angle lens underwater, creating a perspective view no human eye could ever have seen and the illusion that the range of visibility was far greater than it really was. We distort when we use digital settings (or film emulsions like Velvia or Provia) that artificially intensify color saturation or selectively boost certain colors, creating stunning hues that no human eye ever saw. We cheat whenever we put a tinted filter over the lens or fire a strobe on a reef at 40 meters where no red and yellow wavelengths of natural light have ever penetrated.

My intent here was simply to try to salvage a badly flawed image of a subject I'm not likely to have the opportunity to shoot again. If the final result is acceptable to other viewers and closer to what I subjectively remember seeing when I took the picture than what I started with, then I feel I have been successful.

All the pixel data in the final mola image came from the camera – I never painted in a pixel that wasn't in the original image. No selections were made except in the final step before printing, when I used a mask to select the background water before applying Gaussian Blur to smooth out the noise in the water. Each step in the color adjustment process was applied uniformly to the entire frame. Nor was it ever necessary to resort to dodging, burning, selective color replacement or selective desaturation.

Professional color technicians have used techniques like channel mixing – or to use the pre-digital term “plate blending” – to prepare images for publication for many years. There are non-digital analogues for every step I've described in the treatment of this image, all of which used to be done in color process wet darkrooms and pre-press houses. In theory, one could even design an exotic film emulsion (or build extreme curves into the image processor of a CCD) that would do everything I've done to this image in the film or inside the camera, every time.

Color casts as extreme as the example I have used here are rare when taking pictures on land, but are frequently encountered in underwater photography. Channel mixing, or plate blending, can be a useful tool to correct this problem, and in some ways it is actually easier when the red channel is a complete loss than in cases where important pixel data must be protected in all three channels. Nonetheless, plate blending can be usefully applied to a broad range of images, including images with important information in the original red channel, though the techniques involved can be somewhat more complicated.

- Robert Delfs

__________

There are new books on Photoshop for photographers published every year, but the best sourcebook I know – which also offers the clearest explanation of color correction in Photoshop that I have seen – is Dan Margulis' Professional Photoshop: The Classic Guide to Color Correction. The book is regularly revised to accommodate new developments in Adobe's evolving program, so get the latest edition.