I was playing about and got the shot bellow. Notice the strange flare at the bottom ?
I reproduced the issue with EF24-105l at 3 different focal lengths, with EF50/1.4, EF70-300, Zeis Planar T* 50 .
If I shoot the same scene in portrait or upside down I see no flare.
Same problem with shooting through live view (all 3 "silent shoot" modes)
All shots where taken with lens hoods on.
Same shot with my 40D shows no such problems.
To reproduce the problem I have to take any bright light/object on an overall dark scene a bit bellow the viewing area. How far bellow does not appear to be critical.
This is not lens flare.
Took two pictures (a lot more actually, 11 pairs to be precise, I just present one pair here).
All settings were the same for both shots:
The second was with the camera rotated by 180 degrees. Bellow I post one strip of each one (Sorry for size, 1:1 pixel ratio).
Camera on tripod.
Manual focus with live view (focused only once per picture pair).
Release with remote control during live view (see mirror lockup).
Manual exposure (from exif: f4.5, 1/400, ISO 400, EF24-105L @ 105mm).
Shot in Raw and processed in DPP with default settings.
Exported to TIFF and cropped/combined in CS3.
Exported to jpg with maximum quality.
I had to crop a bit to the left/right to make the images best fit.
Observe how the fairly bright sky affects the image quality when it is towards the bottom of the camera.
There is no motion blur, just degradation of contrast.
Check the letters on the solar collectors.
Note how the situation reverses as you go from right to left. Top strip is worse at right side, bottom strip is worse at left side
(do not forget camera was in portrait orientation). The side that was closer to the bottom side of the camera suffers the most.
The top strip has the effect more obvious because the sky at the right side was brighter.
Also note that this is not extreme shooting conditions; the sky is within dynamic range without any highlights recovery.
- I have confirmed that this is due to a very reflective back of the mirror flap.
- It shows on dark frames with strong lights outside the frame (a lot)
- It degrades image quality on every shot to a varying degree unless the surrouning of the subject is black (bottom side).
- It is more obvious if there is backlight (even modest) behind the main subject and at the bottom of it.
- At least one other Canon camera that I tried (40D) does not have it.
Large prints (exchibition stands for me) reveal this issue.
It is psychologically very annoying because the fix is so cheap and easy (yet impossible to do one by himself) and it
degrades performance of high quality expensive gear.
It affects a "landscape camera" like 5DmkII in the worst possible way.
Imagine shooting a landscape with sunny snow and a shift lens for stitching three frames. The large image circle of the lens
and the high brightness of the snow (outside the frame) degrades contrast and color in the frame.
Same goes for many scenarios like water highlights night lights, bright skies etc.
Go buy the best sharpest glass, close the aperture down to the sweet spot, the loss is still there.
This is another test I set up to verify that the problem comes from the mirror assembly.
The idea is to light only one area of the mirror box at a time and check the reflection in the sensor.
Since we have a different behavior for the top and bottom parts we should be able to see the difference.
I did this test with 5DmkII (which did exhibit the problem) and 40D which did not. This further comparison could verify the whole theory.
I shot one camera with the other. The two pairs of shots were with manual exposure (same settings) and direct sunlight hitting
only the mirror backside or the bottom plate with the AF sensor.
The reflection can be easily seen on all 4 but with different brightness on each shot.
The ideal material would have a black hole where the reflection is. In our case the less bright the better.
Note that the lit mirror back on the 5DmkII reflects so much light that almost blows the highlights.
But most important check the reflection in the sensor area. I think the image speaks for itself and explains the whole phenomenon.
The objective is this :
- How much of the image circle hits the 40D mirror back ?
- How much the 5DII back ?
and most important
- Why is the reflection so elongated ? For a point light source coming from an angle close to 90 degrees in relation to the sensor,
it should be very close to circular (after been bounced off the mirror assembly)
I did a very basic modeling of the camera mirror boxes and also a test model of the mirror's back face.
The dimensions are as close as I could measure them.
For the image circle scenes:
-The light source is positioned at the registration distance of the EF mount(44mm).
-It more closely resembles shooting with very small aperture.
-The light projection was set to the minimum required to cover the FF size sensor.
For the reflection model, I set three lights shooting very narrow beams on three surfaces.
The first is a close approximation to the mirror's back housing with the distinctive horizontal grooves.
The second is a flat surface. This should definitely produce an almost-circular reflection.
The third is the same mirror back but rotated 90 degrees so the grooves are now vertical.
The extra two surfaces serve as a test/verification for the ray tracing accuracy.
Here are the results :
Please note that the bottom and right sides of the mirror box are not included so we can see "inside" easier.
- The modeling is capable of resolving reflection, refraction, diffraction and caustics.
Recognize the shape of the reflection ?
Observe how the flat surface produces a circular pattern.
The vertical grooves produce a totally different pattern. The non uniformity is because the grooves are not symmetrical.
It takes the horizontal grooves to get what we see in reality.
And here's some more pics: