James Stephens

Since the introduction of the Canon EOS 5D last year, there’s been a surge in full frame usage and a lot of subsequent discussion about vignetting with various lenses on the 5D in forums, newsgroups etc. Some people are very upset at the vignetting they are seeing. I’ve been doing some thinking on a way to correct the problem …

What is vignetting?

Vignetting refers to a reduction in image brightness in the image periphery compared to the image center. This can be caused by a physical obstruction such as the rim of a lens filter, or optically by properties of the lens itself.

Optical vignetting is also known as light falloff. It obeys a cos4(θ) law, where the light falloff is proportional to the fourth power of the cosine of θ (where θ is the off axis angle).

Is it a problem?

Vignetting is certainly more significant with the same lens on a full frame body vs an APS-C body.

My experience in using the Canon EOS 5D in combination with the Canon 24-105 f/4L IS USM is that I see quite significant vignetting at the wide end of the focal length range and at large apertures. If you look at the full frame vignetting tests here you will see vignetting is in fact present at larger apertures across the entre focal length of the lens. It is unclear exactly which body was used for these tests; the author states it was the highest resolution available so this probably means the 16.7 megapixel EOS-1Ds Mark II.

How can vignetting be removed?

Vignetting can mostly be removed manually in Adobe Camera RAW. There are two sliders for controlling vignetting under the lens tab, one for Amount and one for Midpoint.

Alternatively one may use a plugin, such as PTLens. The same two sliders (Amount and Midpoint) exist in PTLens as for Adobe Camera RAW.

Adobe Camera RAW has the advantage that it operates on the RAW file as opposed to the image data in Photoshop.

Either way though, it can be a time consuming operation to adjust the amount/midpoint levels anywhere close to correctly. One has to be careful not to overcorrect the vignetting as this can result in blown highlights.

A better way to correct for optical vignetting?

A technique known as flat-field correction is commonly used to remove artifacts that arise from varying sensitivity of a detector in a telescope. An image of a uniformly lit white surface, a so-called flat-field image, is used to correct the master image.

It struck me that I’d seen something very similar from my Expodisc white balance filter, and that it might be just the trick to provide such a flat field image. Here is an example from a Canon 5D + 24-105 lens combo at 75mm, f4:

Then I thought .. why not take a series of shots through the Expodisc at varying apertures and focal lengths, correct those in Adobe Camera RAW, tabulate the values and then refer to those values when correcting actual RAW files for optical vignetting?

I tinkered around with the RAW file from the image above in Adobe Camera RAW, adjusting vignetting amount/midpoint by eye, but it turned out to be quite tricky to judge the right levels even with such a featureless image. Then I struck on the idea of using the Color Sample Tool to measure brightness at various points on the RAW image and adjust until they were level. This worked well. For 75mm at f4 I had my Amount/Midpoint values. Time to work out some more.

Then I thought .. this procedure is quite straightforward, surely someone has done this already and could save me the trouble of computing all these values? A search on Google revealed that this wasn’t the case, but I did find this article from someone using exactly the same technique but with a couple of additional tweaks on top of what I had come up with. Before proceeding I need to determine that an Expodisc doesn’t cause physical vignetting on the lens in question - if it does then a better general way of making a flat field measurement will be required. Other than that, this nicely written page lays down the methodology I plan to use to compute amount/midpoint corrections for my Canon 24-105 f/4 IS lens.

If others do the same we could build a useful database of corrections for various lenses on the 5D and I would be happy to host these corrections. Watch this space ….

Update 03/17/06

I discovered tonight that my Expodisc causes physical vignetting on the Canon 24-104 f/4L IS lens. What I did was to take a shot of a white board at 24mm and wide open (f4); in subsequently trying to remove the vignetting on the RAW file I cranked up the Amount slider to 100 and it wasn’t enough.

I also took a shot of the white board without the Expodisc in order to (a) compare this to the shot with the Expodisc to look for evidence of physical vignetting and (b) to see if the board by itself would allow me to get a flat-field measurement. Unfortunately, I wasn’t getting even vignetting in all corners without the Expodisc as the board was at an angle to the light and not lit as evenly as I first thought.

Back to the drawing board for now ….

Update 03/27/06

The most important thing I’ve learnt about vignetting over the past ten days is that it is quite dependent on the strength of the light source. I made numerous attempts to photograph a white board in various lighting conditions in order to obtain a flat field measurement but I could never get the lighting quite right and always ended up with uneven vignetting. I realised later that the implication of this was that even if I could obtain a flat field measurment and thus work out Amount/Midpoint corrections for a given lens focal length and aperture combination then they would only be valid at a specific light level.

It’s not all bad news ….

I was pointed in the direction of DxO Optics Pro 3.5 and after downloading a demo of the software to my astonishment I discovered that it does an outstanding job of correcting for vignetting, as well as other lens distortions such as pincushion/barrel and chromatic aberration. DxO Optics Pro does such a god job because the corrections it applies are based on real measurements of particular lens/body combinations; the corrections are reverse engineered from these measurements. You can read more about this software here.