EOS 5D vs. EOS film camera: any test with wide angle lenses?

[aurelio]

I would like to see a comparison test between:

 

1) an EOS 5D photo taken with a wide angle lens at a stated aperture

(e.g. EF 20mm/2.8 @ 2.8 - 4 - 5.6 - etc.)

 

2) the SAME photo taken with the SAME lens and apertures on a film

EOS, and scanned at the same size of the 5D photo

 

I would like to see if there is any difference in border definition

and vignetting, thus deciding if it is the lens or the sensor ...

 

I think such comparison would be very interesting but didn't find

one on the web until now

 

Perhaps some forum friends have got a 5D and a good EOS film body

and could make such test, which imho would really be VERY interesting

[oofoto]

We already know there is light fall off at the edges on full frame digital sensors - this is not up for debate.

[james_symington1]

Not quite what Aurelio asked but nevertheless a decent point to make are these 2 shots I took seconds apart yesterday. It was with a 5D and the 24-105mm with exposure compensation set to -2/3 to accentuate corner darkening. One shot taken at f4 and the other at f8. At f8 vignetting is essentially absent.

 

Ciao

 

James<div></div>

[roger pfister]

What you will learn (mainly) is how good or bad the scan is/was of the neg (or slide). You might learn a bit about how well the film had been held flat in that film body. And how the particular PP of that 5D shot compared with that film make.

 

What else are you expecting to learn? We know just how good "good shots" from a 5D can be, so what are you expecting to learn?

 

--

Roger

[w_t1]

What Roger said, with a whole bunch more snide comments piled on.

[fourfa]

I think this is a great idea, since there's so much chatter about microlenses and acceptance angles etc. I'll gladly do it if someone donates $100 for the drum scan. Anyone? Bueller?

[scott_eaton]

Ohhh...Ohhhhh! (horshack raises hand).

 

I've a better one. Let's try a 24-70L on the 5D at the wide end and at F4 -vs- a Canon 28mm fixed, or cheap Canon lens kit.

 

Now take the money difference between the two lenses, and buy a film scanner for the second test :-)

[daniel_taylor]

"We already know there is light fall off at the edges on full frame digital sensors - this is not up for debate."

 

We already know there is light fall off at the edges on 35mm film - THIS is not up for debate.

 

What IS up for debate is whether or not the light fall off is worse with full frame digital sensors than with 35mm film. I'll admit to not having tried a direct test. But every image I've seen referenced as an example of "digital" light fall off was no worse than slides from my film days that clearly had vignetting from a zoom wide open at the wide end. And guess what? Seems the "digital" examples also tend to be shot using zooms wide open at the wide end.

 

Who would have thought?

[denisbergeron]

Since both camera have same sensor size (one electro-static the other chimico-physic), both will have the same vignetting.

For the difference between film and digital, even if it's not the same lens / company, you can find a good comparision there :

http://www.ales.litomisky.com/shootout/analogversusdigitalshootout.htm

 

You can rent both EOS film camera and EOS 5d in a store near you and compare it !

[lad_lueck]

I shoot the 24-70 on a FF 1Ds. Falloff on *test* shots at 24mm/2.8 is minimal, and completely cureable with Photoshop or Adobe Camera Raw processing.

 

If you are using this criteria to decide whether to go FF, you are not even close to worrying about important factors... IMNSHO!

 

*Why **test** shots?* What subject would you realistically shoot like that?!

[digitmstr]

>>"We already know there is light fall off at the edges on full frame digital sensors - this is not up for debate."<<

<p>

ACtually, "we" do NOT know. There is no conclusive proof, only rumors. On the other hand I have a track load of images taken with my 5D and various lenses from 15mm to 200mm that are indeed proof that light fall off on the 5D is negligible and no different than what I used to see with the same lenses on my EOS3 and 1v.

<p>

The sample posted taken with the 24-105 is not a light fall off issue. If it were, the "vignetting" would be less at the smaller aperture and not viceversa. I have never seen a lens vignette more at smaller apertures!

<p>

<a href=http://www.vanwalree.com/optics/vignetting.html>Here's</a> a very good text on the subject of light fall off.

[mark u]

Actually, we know a fair amount about this issue. It was given widespread publicity by Chasseur d'Images in their March, 2003 issue where they tested a number of lenses on a 1Ds, and observed both additional vignetting and colour shifts with wide angles and fast apertures. They reported their results as data, rather than images, so it is perhaps a little difficult to visualise what is going on - either optically, or in terms of the actual impact on images.

 

van Walree's article on vignetting is excellent (and should be understood by anyone trying to assess the imact of sensor optics) - indeed his whole Optics section of his site offers the best explanations I have encountered on the internet for the various topics he considers. Unfortunately, he has yet to offer similar clarity with sensor optics in comparison with film.

 

I have made some simple ray trace charts in Excel to show what happens with microlenses. A microlens is basically a hemispherical lens of radius about half the pixel pitch (3-4 microns) that covers the pixel - it sits atop the Bayer colour filter, which is typically about 1 micron thick and bonded to the pixel silicon. For simplicity, I have assumed that the refractive index of the lens, CFA and silicon are all the same (1.53 - fairly typical for crown glass) - minor variations won't change the optical path significantly. Light is typically absorbed after passing through different path lengths of silicon, depending on its wavelength. Blue is absorbed close to the surface - 0.2 microns, green averages 0.6 microns, and red averages about 3 microns (although some red photons may not be absorbed at twice that depth). The first chart shows ray traces for light at an angle of 10 degrees off vertical.<div></div>

[mark u]

As can be seen, the rays are focussed within a relatively small area of the pixel, although there is a high degree of spherical aberration. The next chart chows the ray trace for an angle of 40 degrees off vertical (about the practical limit, given the nature of the EF lens mount). These rays come to a focus close to the periphery of the pixel (and probably outside the photosensitive area), and the path length through the silicon is much more variable, leading to different degrees of absorption for red wavelengths compared with the near vertical case:<div></div>

[mark u]

The above diagrams should make it easy to understand why Leica opted to make microlenses that are increasingly offset like shift lenses the further a pixel is from the optical axis of the lens. It should also be easy to appreciate that the effect is to cause increased vignetting toward the image edges with lenses whose exit pupil is closer to the sensor plane. With a fast lens, even in the image centre there are rays that are focussed over a wide cone, so the microlens vignetting effects (which affect different wavelengths to different degrees, as discussed above, causing colour shifts) apply.

 

Of course, it is possible to attempt to correct for these effects in software or firmware - as is done with the Olympus 4/3rds cameras. Canon has not publicy said anything about what steps they have taken to deal with these problems AFAIK, but the effects are quite real and an inevitable consequence of simple optics and the physics of light and of silicon. The simple analysis above is insufficient to make any quantitative conclusions, which would require intensive calculations based on the optics of specific lenses - however, the qualitative issue is unequivocal.

[berg_na]

The root cause of the corner vignetting in a CMOS sensor stems from the fact that the photosensitive area of the pixels lies several microns below the surface of the chip, therefore, part of the incident light at oblique angles is blocked off from the off-axis photodiodes (making them appear darker). This effect is fully understood, and well documented.<div></div>

[mark u]

That is largely a myth. As I stated in my original post, blue and green are absorbed quite close to the silicon surface, while red wavelengths are absorbed more deeply. See this paper from Foveon - in particular pages 2 and 3 which show the relationship (and the structure of a Foveon pixel):

 

http://www.foveon.com/files/CIC10_Lyon_Hubel_FINAL.pdf

 

The chart showing absorption vs. depth of silicon on page 3 is here:<div></div>

[berg_na]

A myth? I don't think so <a href="http://white.stanford.edu/~brian/papers/pdc/opticalefficiency.pdf" target=blank>(Link)</a>... The varying absorption depth in silicon at different wavelengths is not the cause of the problem, this property is completely independent of the pixel location, and the corner shadowing affects all colors.

[mark u]

Nice paper - but it relates to ancient technology by DSLR sensor standards - indeed to an era when CMOS was not considered viable for quality sensors. The pixel structure they analyse was imaged in 1999 and used 350nm technology (see reference 5 in the paper), and is purposely chosen to have a buried photodiode to demonstrate their analytical capabilities - it appears to have been built by either Agilent or Philips, whose sensor business became DALSA, IIRC. They make reference to a very low fill factor (just 30%, largely the result of the 350nm line size), and no reference whatever to microlenses.

 

Here is what DALSA say about their current sensors:

 

"Fill factor: the fill factor of our sensors is typically 80-90%."

 

"Better angular response: DALSA sensors provide excellent angular response, a characteristic that is very important when using fast lenses and wide angles. We offer thinner polysilicon layers and fewer structures above the charge-collection region of the pixel to reflect and distort incident light, and with our high fill factor, DALSA sensors generally do not require microlenses. Microlenses help increase effective fill factor in some sensors, but they make the performance of the pixels highly dependent on the angle of incoming light rays. Microlensed pixels do not respond uniformly at larger iris openings (low f-numbers), particularly at at the edges of the sensor."

 

Of course, DALSA's pixels are slightly larger at 9 microns than the 5D's 8.2 micron pitch, but Canon's pixel technology is probably superior at a given pixel size (they have a much larger R&D budget), even though they do use microlenses.

 

Where I suspect the broad conclusions of the Stanford paper remain important is for digicams with small pixels - some of which are smaller than 2.5 micron pitch, and so have an area less than 10% of a 5D pixel. This would also account for the rarity of lenses with an angle of view wider than about a 38mm lens in 35mm equivalent terms on those cameras.

[berg_na]

Unfortunately, this 'ancient' issue is still present in all of the current CMOS sensors. You're absolutely correct in saying that it will be more pronounced in smaller pixels, but it clearly has an impact on the 5D sensor as evidenced by the corner vignetting.<p> The "80-90 percent fill factor" <a href="http://www.dalsa.com/pi/products/DSC.asp" target=blank>comment from Dalsa</a> refers to their CCDs, not CMOS sensors, in fact, the next sentence reads <i>"CMOS sensors have much lower fill factors due to the extra structures or circuits they place on every pixel".</i> All current CMOS sensors rely on a photodiode to convert photons to electrons, and the photodiode in each pixel sits at the bottom of a tunnel.<p> Note that the overlying layers are primarily aluminum wires, not polysilicon, Dalsa's comment about using thinner polysilicon gates again refers to the CCD technology where polysilicon gate electrodes cover each pixel to capture and transfer the photocharge packets.

[eric merrill]

So the 5D probably vignettes slightly more than a film camera.

 

Big whoop. Who cares?

 

If you need to shoot film, it doesn't matter how the 5D performs.

 

If you need to shoot digital, it doesn't matter how the film camera performs.

 

If you don't *need* digital, then don't buy yet. There will be a better and cheaper choice next year.

 

For my money, the 5D performs very well wide open with wide angles. I'm extremely happy with the 35/1.4 at 1.4.

 

Is there light falloff in the corners? Yes. Do you notice that when you're shooting in a dark room and most everything is dark anyway? No. Do you notice that when you're shooting a white wall with lots of light and you don't really need to be at 1.4 anyway? Yes, but what's the point? :)

 

 

Eric

[digitmstr]

The theory of angle of incidence is understandable. Its practical effect are not proven by a long shot. Every test I have seen was done in a lab hitting the sensor at angles which would never occur in real life, using real lenses.

 

Doing a test with the same lense on both film and DSRL (making prints, NOT film scans) may reveal some differences. Although, those would be difficult to measure because the print from film and digital would have to have the same exact exposure, etc...

 

In the end, much ado about nothing. Coming from OES3/1v to 10D and 5D I noticed NO difference whatsoever in ligh fall off using the same lenses I had on the the film body.

[fourfa]

how about this. same lens on 5D and EOS3 (slide film), tripod, pointing at a centered macbeth colorchecker on an evenly illuminated white wall. spot meter on the middle grey patch. develop and scan (Coolscan V, FH-3 film holder which is very good for flatness), set the levels in photoshop so the black and white and middle grey patches match in RGB values for both images. Then sample RGB values in a radial line out from the center to measure fall off.

 

I don't have an opinion about what will happen, and I don't really care (as long as I have the vignetting tab in ACR, but I do have a pet peeve about armchair photography when experimentation is so easy to do