Vignetting on Full-frame sensor quantified (with 2 lenses)

[Bart_van_der_Wolf]

As my first post here I'd like to share my findings on the vignetting/light fall-off that I've quantified for two of my lenses (others may follow) on the EOS-1Ds Mark II. I've not done a comparison with a film image of the same size, but I've no a priori reason to assume it's much different. To me it seems that the microlenses do a good job at decreasing the angle of incidence. I've also chosen to post here instead of in a more lens related forum, because the results can/will differ between sensors, and I wanted to avoid a 1.6 crop versus FF sensor debate as much as possible.

I used Imatest to take the tedium out of the testing procedure, and summarized the results in a graph. To provide a uniform 'flat'-field for exposure, I used a piece of opaline glass held flush to the front of lens, which was pointed at the sky (overcast), neither filter nor hood was used. The exposure was set to +1.0 stop on Aperture priority. The Raw file was processed as a linear 16-b/ch RGB, and the darkest corner was plotted in the summary (as a worst case scenario).

Here is an example of one of the Imatest output options for an f/8 aperture, and this is the summary for the full range of apertures of my EF 50mm f/1.4 .

What may be interesting for those seeking to reduce vignetting/light fall-off is, that stopping down beyond f/4.0 serves no purpose (other than for DOF and diffraction), it won't get any better stopping down further.

As a comparison, I've also verified my findings with another lens of somewhat similar focal length, the T/S-E 45mm f/2.8 in an unshifted state. It also won't get any better beyond a given aperture, f/4.5 in this case (which happens to be where the lens also peeks in resolution). Another noteworthy observation, the wider image circle won't change the source of the phenomenon.

What happens is, that at the widest apertures the cone of light at the exit pupil shows signs of the barrel construction partially cutting-off the most oblique rays and the cone going from an elliptical diameter to a more round one at smaller apertures. Beyond that, it's mostly the larger distance to/magnification at the corner that causes the loss of intensity. That law of physics is a given, it will be darker at the corner (with a rectilinear projection that is).

Also, These 'worst corners' are indeed that, the most extreme corners of the sensor, and things already improve when you move in a millimetre, so don't despair ..., but be prepared for the additional fall-off that the shift will bring, but that's something for another subject (and forum I guess).

Bart

[Diane Fields]

I thought I would add this link since I had just read it on another forum (and its been linked to from yet another). I do better with images and maybe some others do also. Though these aren't exactly the same lenses you used, it gives one a good idea also.
http://www.photo.net/bboard/q-and-a-...g=200606111006

[Tom Wilk]

Good information, Bart. This issue does seem to crop (no pun intended) up frequently among the various forums. It's always good to see some quantifiable data on vignetting.

Also, good link, Diane - I saw that one yesterday and found it very interesting as it does a good job of illustrating the strong similarity between full-frame digital and 35 mm film in terms of light falloff (vignetting).

[Peter Facey]

I' m particularly pleased to see Bart's graphs for vignetting because, when I first bought my Canon 5D in November 2005, I did the same Imatest light falloff tests on two of my own lenses and was horrified by the results. My graphs, which are upside down compared to Bart's, are very similar to his. Mine are here.

Now, two years on, I should add that in practice, with real landscape photos, I hardly ever feel the need to remove the vignetting using Lightroom or DXO, so what looks horrid in a scientific test is mainly quite ok for normal photography.

The 1DsII is of similar vintage to the 5D. It will be intersting to see if improved microlenses or fill factors in the 1DsIII or D3 improve matters.

Also, many web reviewers report vignetting incorrectly. Compare for example slrgear's graph with mine. I'm not sure why this is, but if people analyse Canon JPGs without even noting what Picture Style they used, it may be that their analysis fails to reverse out the effect of the JPG tone curve's gamma, unlike Imatest (in Bart's case he was using linear conversion from raw so gamma was 1).

[Asher Kelman]

Hi Peter,

I'm impressed with the Imatest software. This program does work that cost a fortune using sophisticated Zeiss lens analysis equipment. What made you go to the extent of using it yourself? Are you perhaps an astrpohysicist or maybe you need flast illumination for architecture or measurement of objects in microscopy?

For artistic photography, the vignetting observed is part of what makes pictures work!

Asher

[Bart_van_der_Wolf]

Originally Posted by Asher Kelman

For artistic photography, the vignetting observed is part of what makes pictures work!

Understanding one's tools is part of mastering them, rather than having the tools mastering its user!

Some obsess about vignetting or light fall-off on (mainly full-frame) sensor arrays, but it is not too much of an issue on retrofocus designed lenses. The angle of incidence and the microlenses will make sure that the lightrays strike the silicon rather perpendicular to its surface. The maximum angle of incidence is restricted by the physical lens mount diameter, and the lens design (given the fixed lensmount to film distance).

On most '35mm' camera designs the vignetting issue is actually probably less significant (due to microlenses) than on film. On some medium format designs the light fall-off will cause color shifts, requiring a software 'lens cast' calibration to reverse the effect.

Once we recognize the effects of vignetting (lens shading) and light fall-off (due to geometrical/magnification effects) in the gamma corrected image, and as shown possibly affected by the chosen aperture, we can also control the effect and use it to our advantage. One can, for artistic purposes, amplify the darkening of the corners to draw more attention to the center of the image, or one can inverse the effect to the point of lightening the corners, e.g. in a high-key image, to achieve the same effect. Most images however benefit from a neutralized to slight vignetting effect, IMO.

Bart

[Peter Facey]

Originally Posted by Asher Kelman

Hi Peter,

What made you go to the extent of using it yourself? Are you perhaps an astrpohysicist or maybe you need flast illumination for architecture or measurement of objects in microscopy?

- No, I'm just an amateur photographer in the UK, but with a degree in physics, so I saw Imatest as a fun thing to play with and combine both interests. (This is an "imaging technology" forum!) And I agree with Bart that "Understanding one's tools is part of mastering them". Whereas some people can learn by doing, I find if I cannot understand something I can't use it well.

Back in 2005, I actually wondered if Imatest was miscalculating its results, but investigation suggested it wasn't. I'm just happy to see Bart got similar worst-case figures on different lenses.

[Bart_van_der_Wolf]

Originally Posted by Peter Facey

I'm just happy to see Bart got similar worst-case figures on different lenses.

Yes, but do remember that the worst corner results already improve when you e.g. get a millimetre away from the corners, let alone from the edges. And again, part of it is due to physics (corner magnification due to projection on a flat surface), and it will be the same for similar constructions (lens mount diameter). It's the vignetting by the tubular construction of the lens that varies.

Bart