Digital Photography For What It's Worth

Red-eye control—an ounce of prevention is indeed worth a pound of cure

On this page—

• An Ounce of Prevention...

• A Pound of Cure...

• Reflex Colors

• References

Last updated July 27, 2004

An Ounce of Prevention...

Red-eye and its four-footed variants blue-eye, yellow-eye and green-eye (the last kindly demonstrated above by our young border collie Megabite) amount to nothing more than unwanted flash reflections bouncing back into the camera lens from the subject's retinas.

A single red-eye session with any photo editor should convince you that an ounce of red-eye prevention is truly worth a pound of post-processing cure, especially when there are several images to fix. Exorcizing red-eye after the fact is sure to take more time and effort—often with dubious results. 

The Reflex

In theory, red-eye shouldn't happen. In an optically ideal eye with completely reversible lightpaths, light from the flash would reflect back to the flash; ring flashes excepted, perhaps, none would enter the camera lens.

Optical marvels that they are, however, mammalian eyes are imperfect retroreflectors. Optical aberrations along the path through the eye cause flash reflected from the retina to emerge from the pupil in a cone beam whose angular width is controlled in large part by the diameter of the pupil at the time of exposure. To simplify the discussion, I'll refer to this return beam of flash as the reflex.

All the pre-exposure red-eye countermeasures discussed here suppress red-eye by either ducking, narrowing, dimming or redirecting the reflex cone.

The Short Answer To Red-eye Reduction

To mitigate red-eye before the fact:

For best results, combine these countermeasures as best you can.

Reflex Cone Width vs. Lens-Flash Angular Separation

The angular width of the reflex cone is controlled at least in part by the angular size of the pupil as seen from the retina. If the angular separation between flash and lens as measured at the subject can be made greater than the angular width of the reflex cone, there'll be no red-eye.

A typical 25 mm adult iris-retina distance and a 3 mm radius for a maximally dilated pupil amount to a worst-case reflex cone half-angle of ~ 7° = arctan(3/25). I suspect that the eye's convergent lens-cornea optics would act to narrow the reflex cone even more, so 5° is probably a good minimum width estimate in most circumstances; 10° is probably an overestimate. I haven't field-tested these calculations quantitatively, but they seem to be in the ballpark in my experience.

The strategy of keeping flash-lens angular separation wider than the reflex cone jives with the common observation that red-eye's a much bigger problem at longer camera-subject distances: To maintain the required minimum 5 degrees of angular separation between the flash-retina and retina-lens lightpaths, lens-flash distance must increase in direct proportion to the camera-subject distance.

Taken to extremes, however, increasing flash-lens separation may introduce unwanted shadows.

Note that flash position around the lens makes no difference with regard to red-eye due to the cylindrical symmetry of the flash-retina system. You'll have no better  luck with your flash at 2 o'clock relative to your lens than you will at 10 'o'clock or 6 o'clock.

Red-eye is a form of glare off a non-metallic surface. As such, it might in principal be subject to mitigation with a polarizing filter. However, the narrow angle of the reflex cone makes me think that a polarizer would be ineffective at any orientation. If you've had success in reducing red-eye with a polarizer, I'd love to hear about at dpFWIW@cliffshade.com.

Estimating 5° Angles

Luckily for us photographers, the tangent of 5 degrees is very close to 1/12, so 6 feet of camera-subject distance translates into 6 inches of minimum lens-flash separation to place the lens outside the reflex cone. Same with 10 feet, 10 inches, and so on. To go out to 10° of lens-flash separation, it's still a good approximation to double the camera-subject distance in feet to get lens-flash separation in inches.

Learn to estimate a 5° angle. At the end of my arm, the knuckles of my long and index fingers together subtend an angle of about 5°.

A Pound of Cure...

You'll find that it's generally much less time-consuming and frustrating to prevent red-eye than it is to exorcise it from your photos after the fact, even with the best of tools and techiques. You'll also generally end up with better results. But when red-eye manages to sneak past all your defenses, post-processing becomes your only way out. 

Many popular photo editors—Adobe PhotoDeluxe, Microsoft Picture It! and PhotoDraw2000 among them—have semi-automated anti-red-eye features, but these generally work no better or faster than the manual PhotoShop methods described below and often create more problems than they solve, at least in my hands.

Photographer Chuck Ross's posted this simple, quick and effective PhotoShop-based red-eye fix on RPD:

While Photoshop doesn't exactly bill it as "red-eye removal", here's what I do, altho there are many ways to accomplish red-eye removal:

Using the Magic Wand, select both red areas in each eye at the same time. Then go in the "Image" menu to "Adjust->Channel Mixer"... [and] lower the red content in your selection till it looks right from 100% down to whatever it takes.

This is FAR preferable to using a color to "paint" the eye, since that will completely distort the detail in the iris of the eye and will look very fake.

I've had success with Chuck's method in PhotoShop LE.

I haven't tried this one from a different Chuck, but it sounds promising:

Zoom in and use Magic Wand to select the redeye area. Take the color sampler tool and grab any pixel you can of the nearest correct eye color. Fill it. Now reselect the [pupil] and use Gaussian Blur.

TECHLAB posted this RPD reply:

>  ... I'm tired of going thru multiple steps in photoshop. 
>  I'm looking for a plug-in or a software package were you
>  can remove red-eye in a quick couple of clicks.

Frankly, I don't think you'd really want one.

Any software that does this would be likely to mistake other things for red-eye and unnecessarily degrade your image.

Since you're using Photoshop, then I can assume you want to keep the image as high quality as possible...

If the image warrants it, I use the Sponge tool in Desaturate mode. This eliminates the color altogether. Red-eye is gone. But, the image has to have a high enough pixel count to allow you to do this. Not all images are appropriate for this technique. 

Another one is to replace the pixels with the Rubber Stamp. Take a sample from a 'good' area and cover the red with it. Use a small enough brush and it's just a few clicks.

Some people use the Magic Wand to select the red pixels and then either desaturate them of fill them with another color. It works, but personally I find it to be a bit too much for MY work. It might select other pixels of similar color in the image....

---

Finally, here's Juri Munkki's PhotoShop technique from RPD:

I came up with a really good way to eliminate red eyes from flash photos. It's fast and the results I get are very natural:
 
        Go to selection mask painting mode
        Paint over the red parts of the eyes
        Invert
        Go back to the regular selection mode
        Hit Command-3 to work on the blue channel only
        Copy the selection to the clipboard (Command-C)
        Hit Command-1 to work on the red channel only
        Paste
        Set the floating selection opacity slider to about 75%
        Hit Command-0 to view the full RGB results
 
Everything except the first two steps could probably be combined into an action ... The idea is to copy most of the data from the blue channel. Works better than anything else I have ever tried.

See, red-eye prevention really is worth a pound of cure.

Reflex Colors

The consistently red color of the human reflex derives from the red blood pigment hemoglobin. Light from the flash picks up the red from blood vessels encountered during its bounce off the retina, just as reflected sunlight picks up the color of a red sweater.  

Why, then, do animal reflexes come in so many other colors and seldom in red? The answer lies in the tapetum lucidum, a highly reflective, variably pigmented membrane backing the retina in animals with good night vision (including dogs, cats and most domestic animals) but entirely absent in humans. The tapetum lies directly behind the retinal photoreceptors. Nova's The Nocturnal Eye nicely illustrates the anatomy. 

The tapetum enhances low-light vision by giving retinal photoreceptors a 2nd crack at any incoming light that manages to escape absorption (detection) on the first pass. In dogs, at least, an additional boost may come from tapetal fluorescence, which shifts incoming wavelengths into better alignment with the peak spectral sensitivities of the photoreceptors. Tapetal pigments surely come into play here.

When tapetal pigment is present, its color dominates the color of a given animal's reflex. Tapetal color loosely follows coat color. For example, black coats and green reflexes tend to go together, as seen in our border collie above. Most dogs and cats show a blue reflex as their eyes mature in the first 6-8 months of life. Pigment-poor animals like blue point Siamese cats with no tapetal pigmentation show a red reflex for the same reason humans do.

References

(See also the home page links.)

Under construction...

Feynman RP, Leighton RB, Sands M, The Feynman Lectures on Physics, Vol. 1, Addison-Wesley Publishing Company, Reading, MA 1963—for a review of physical optics.

Pickett JR, DVM, "Why do dogs get blue, not red, eyes in flash photos?", Scientific American, Vol. 285, No. 3, p. 104 (September, 2001)

What do dogs see?, North American Hunting Retriever Association (NAHRA), 1996. 

Unless explicitly attributed to another contributor, all content on this site © Jeremy McCreary

Comments and corrections to Jeremy McCreary at dpFWIW@cliffshade.com, but please see here first.