Returned my S90: Tilt / distortion / whatever.

I was quite impressed with what the S90 will do.

If you are seriously going to take these types of shots, Canon has new shift/tilt lenses just for this purpose. Granted the lenses alone cost slightly more than a S90; however, they are marketed for the simple reason that a camera like the S90 isn't designed to provide the best reproduction of scenes of this sort.

SDRebel said:

I was quite impressed with what the S90 will do.

If you are seriously going to take these types of shots, Canon has new shift/tilt > lenses just for this purpose.

Click to expand...

Maybe Canon should have called this the S90/TS-E instead, with the tilt preset at the factory ;-)

It cannot be a "tilted" sensor as the image that you are using to align the camera before taking the shot is also coming from the same sensor.

I have seen this kind of problem with some Canon EF IS (SLR) lenses. When the IS system is engaged, the image can shift by almost a degree. However, these were telephoto lenses (like the 70-200 IS) and I am not sure how sensitive the wide-angle like zoom lens on the S90 would be to a shift due to IS.

Did you frame (align) your shot while the IS system was not active?

In order to avoid geometry distortion with a wide angle lens, you have to be centered both in the horizontal and and vertical direction with respect to the plane you are facing. It is relatively easy to center your camera in the horizontal direction, but centering in the vertical direction is quite hard...

It is hard to judge from the picture your exact location when you took this shot, but it looks like you were a bit higher than the vertical center point of the wall you were facing (behind the bench) which would explain the kind of distortion that is present in the picture.

IS doesn't affect roll, as it can't correct along that axis.

At first, I thought the OP was complaining about the fact that the horizontal border that he aligned his frame to before taking the shot is not perfectly aligned with the horizontal grid in the shot. The picture appears to be tilted right by a fraction of a degree. This is exactly the kind of frame shift I have observed with a couple of EF IS SLR lenses.

However, there are other problems with the picture. Specifically, the camera was neither horizontally nor vertically aligned to the midpoint of the main wall with the picture. This would certainly cause perspective distortion which is inevitable. Latest versions of Photoshop has a tool for fixing this kind of distortion but it is quite labor intensive...

macshark said:

However, there are other problems with the picture. Specifically, the camera was neither horizontally nor vertically aligned to the midpoint of the main wall with the picture. This would certainly cause perspective distortion which is inevitable.

Click to expand...

I know Tom's very thorough in his methodology so I ruled that out. I then proceeded to run my own test, which you can find here:

http://forums.dpreview.com/forums/read.asp?forum=1010&message=33352904

I have used a ruler to measure the distance between each end of the camera and the mirror to ensure it's parallel with it.

I know what you mean with the image shifting because of an engaged IS, but it will shift , not rotate . IS will perform and x/y axis shift of a lens (or group thereof) in order to compensate for the lens movement, but never a roll around the z axis. I notice the shift all the time on my 100-400mm L lens when I half-press the shutter, but never a rotation (tilt).

If it worked it would represent a considerable savings.

Olga Johnson said:

Paul Marbs said:

Well, I have to say the plot thickens!
What's the point in the G11? I can see the point in the S90 but really the G11?
Might as well stick with the LX3 or get a GF1 if you're gonna get a G11..

Click to expand...

1. 1: Articulating LCD
2. 2: zoom to 140mm v. 60mm
3. 3: Color. I have a heck of a time with LX3 color especially when the sea/water is involved. It always need correction of color in Photoshop. With the G11 I can simply resize jpg and be done. LX3 requires more work.
4. 4: External flash - The 270EX flash (which I use with my dSLRs) is perfect for the G11.
5. 5: Controls - The G11 has easier access to ISO changes and white balance (which I assigned to the "S" button.)

Menu with G11 is much more intuitive (at least to me being used to Canon menus).
And I can fit my G11 in the same neoprene case that I used for the LX3.

--
Olga

Click to expand...

I have the LX3 and was wondering what the G11 with the EX270 flash is like for portraits bouncing of ceiling etc, Also how balanced is the combination.

LX3 has a hotshoe but there is no Bounce flash available that does,nt completly dwarf and unbalance the camera(looks stupid as well)

I'm not sure of how do you judge when the picture is straight? I would look on the display, but it seems like you used other methods?
Luc

--
http://www.pbase.com/duca_v2

OM1 said:

Tom Hoots wrote:

I don't think so. From the result of the mirror test that I suggested and you performed with your S90 (see http://forums.dpreview.com/forums/read.asp?forum=1010&message=33304478 ff) it seemed quite apparent to me that the image tilt is indeed due to a tilted sensor.

Click to expand...

Sorry but... come on! Tilted sensor? So the camera was set 100% straight parallel 0 degrees to the wall according to what measurement?

You just look at the LCD and you see what you will get, period. If the camera is tilted, the image is tilted. it's that simple
Luc

--
http://www.pbase.com/duca_v2

Hey Tom. Yes, I assumed you did that, but it was the only thing I could think of as a possible explanation other than a slightly tilted sensor. Good luck on your continued quest!

Tom Hoots said:

BravoBrown said:

Btw, unless you were directly in the center of the room, there would be no auto correction for that. If you were directly in the center, the sensor was slightly tilted, not just lens distortion going on.

Click to expand...

I was as close to center as I could possibly get, as far as that goes....

Tom Hoots
http://thoots.zenfolio.com

Click to expand...

--

I am still a bit confused.

In the OPs case, he aligned his camera using the LCD display.

If you are aligning the shot using the LCD display image, the sensor tilt should be compensated for.

Sensor tilt would be a problem if there was an optical viewfinder or aligning the camera using a level.

If the problem is such that after aligning the subject using the LCD display, the captured image has a different alignment, this would indicate a different type of cause.

I can only speculate as to what may be causing this, but here are some ideas:

1. Live view LCD image doesn't apply lens correction. The captured image has lens correction applied by in camera software. There is some mismatch due to this. One way to check this would be to take a look at the RAW image.

2. Live view LCD image applies a quick and dirty form of the lens correction algorithm that would be applied to the final image. However, due to the speed optimization or a bug, the LCD live view lens correction behaves differently than the final lens correction resulting in a small tilt between the two images.

Tom Hoots said:

Luciano Morpurgo said:

To me it just looks like the camera wasn't exactly parallel to the front wall. You just tilt it a bit to the right, and the distortion increases on the right. The closer is an object, the more it gets distorted.

Click to expand...

Well, I think the "proof of concept" was made during the original series -- wherein, absolutely every time, over and over and over again as I tested the different ISO settings, and remember I take dozens of shots where other folks might take just one -- the LX3 and the WX1 produced pictures that were essentially "straight," whereas every single S90 shot had this "tilt" to it.

Yes, I fully realize that no lens is perfect -- indeed, the LX3 and the WX1 weren't geometrically "perfect" by any means -- but there's an "amount of imperfection" line that the S90 clearly crossed, for the worse. And this last test was probably about as close as anyone could get without getting into "laboratory" conditions -- it's not difficult to find the exact center of the room, given the carpet patterns, the seat positioning and such, and then the angle of the camera is a byproduct of getting into that dead center position, and then I could use the many patterns in the room to make sure I got the shot completely centered.

Click to expand...

Don't know, it doesn't look so simple to me anyway. I would never try to do such a centigrade precision test handheld...You should also assume that the room is 100% straight. And how are you sure to put the camera in the "dead center" position?
Luc

--
http://www.pbase.com/duca_v2

Sye68 said:

I have the LX3 and was wondering what the G11 with the EX270 flash is like for portraits bouncing of ceiling etc, Also how balanced is the combination.

LX3 has a hotshoe but there is no Bounce flash available that does,nt completly dwarf and unbalance the camera(looks stupid as well)

Click to expand...

It's a very balanced combination. The bounce flash works well. I could quickly do these:

In P mode, direct flash with 270EX:



In P mode, bounced flash (high ceiling in room) with 270EX:



--
Olga

macshark said:

I am still a bit confused.

In the OPs case, he aligned his camera using the LCD display.

Click to expand...

He certainly aligned the position and shooting angle of the camera, but probably not its rotation around its optical axis. After all, he wasn't using a tripod but rested the camera on the railing of a balcony.

macshark said:

If you are aligning the shot using the LCD display image, the sensor tilt should be compensated for.

Click to expand...

I guess that the slight tilt (we are talking about less than 1 degree here) was present in the live view as well as in the resulting photo, but just barely visible on the camera's LCD (due to its very small size and very low resolution compared to the displays we are using for our pixel peeping).

Folks,

Thanks for your thoughts. I'll try to respond to a few things that many of you have written.

First of all, in response to the many thoughts that I may not have been perfectly centered in all planes:

Well, you're probably correct. But, as I have maintained, this was probably about as close as anyone is going to get in a "real-world" test. It won't be long before we'll have exacting lab tests that'll show what at least one S90 example looks like.

Still, I just have to point you to my high-ISO test from last week, in which I compared the S90 to the LX3 and the WX1, here:

http://thoots.zenfolio.com/p916892765/h2701ce48#h2701ce48

Just go through those, and see how every single S90 shot appears to be tilted, as compared to the other two cameras. I wasn't trying to be "exacting" at all -- I just placed each camera in the same place as the others, and tried to take shots as "centered" as I could.

The only difference between those shots and the one I have posted above is that I did try to be as exacting as I could in this newer shot.

In the end, I think the "comparison" shots tell the story -- compared to other cameras handled in essentially the same way, the S90 showed an obvious tilt. Yes, the other two cameras were "zoomed in" a bit in an effort to equalize the focal range, but I can show you full-wide examples from those cameras that don't show any kind of a tilt like the S90 did.

So, I could be even more exacting, but I don't think that's necessary. I've had something like a dozen different cameras on that railing, shooting the same test, and this is the first time I've ever seen the kind of geometrical non-linearity as the S90 produced.

In response to the thoughts that "if the sensor was tilted, the on-screen grid display would be tilted, too:"

Agreed. I really don't think this is a "tilted sensor" situation. I just don't see the "tilt" as being uniform across the height of the image -- I see more tilt in the top half of the image than the bottom half. Oh, maybe that has to do with "I wasn't centered enough," but I'll press on, regardless.

Go back to my comparison gallery, and look at all of images for geometrical non-linearity. None of the cameras are "perfect." I don't expect any camera to be "perfect." But there's certainly something "a whole lot more wrong" with the S90 linearity than that of the other two cameras.

I really see what you could call "non-symmetrical non-linearity" in the S90 images, especially in the upper half of the images. I would call it, in the horizontal plane, in the upper half of the S90 images, it's like there is "pincushion distortion" on the left side (corner bending up), and "barrel distortion" on the right side (corner bending down). If my S90 had had "normal barrel distortion" on both sides, I'd probably just call it "more lens distortion than I'd like to see."

In the end, "whatever it is," it was way too much "geometrically wrong" than I'm willing to put up with. And I found it to be "easily noticeable," in comparison to all of the other camera I've ever owned.

Tom Hoots
http://thoots.zenfolio.com

Well, in order to buoy the spirits of S90 owners (after reading Tom's critique), take a look in the upper right corner of his iso 400 comparo shots. Read the writing on the wall if you dare. Which camera produces the highest legibility? That would be the S90.

Yes the images of S90 shots are tilted to the right side and it is enough to be noticeable to an unsuspecting eyes. Question is whether it is an isolated phenomenon confined to S90 you happened to test or other S90s are affected too. If latter is the case, I hope certain batches only are affected. Too early to tell because no other similar phenomenon is reported but I think you should make an inquiry to Canon for this problem.

P.S.: Would you confirm this in the following thread? You just need to shoot a telephoto shot or your naked eyes will be enough.

http://forums.dpreview.com/forums/read.asp?forum=1010&message=33348827

Tom Hoots said:

Still, I just have to point you to my high-ISO test from last week, in which I compared the S90 to the LX3 and the WX1, here:

http://thoots.zenfolio.com/p916892765/h2701ce48#h2701ce48

Just go through those, and see how every single S90 shot appears to be tilted, as compared to the other two cameras. I wasn't trying to be "exacting" at all -- I just placed each camera in the same place as the others, and tried to take shots as "centered" as I could.

The only difference between those shots and the one I have posted above is that I did try to be as exacting as I could in this newer shot.

In the end, I think the "comparison" shots tell the story -- compared to other cameras handled in essentially the same way, the S90 showed an obvious tilt. Yes, the other two cameras were "zoomed in" a bit in an effort to equalize the focal range, but I can show you full-wide examples from those cameras that don't show any kind of a tilt like the S90 did.

So, I could be even more exacting, but I don't think that's necessary. I've had something like a dozen different cameras on that railing, shooting the same test, and this is the first time I've ever seen the kind of geometrical non-linearity as the S90 produced.

In response to the thoughts that "if the sensor was tilted, the on-screen grid display would be tilted, too:"

Agreed. I really don't think this is a "tilted sensor" situation. I just don't see the "tilt" as being uniform across the height of the image -- I see more tilt in the top half of the image than the bottom half. Oh, maybe that has to do with "I wasn't centered enough," but I'll press on, regardless.

Go back to my comparison gallery, and look at all of images for geometrical non-linearity. None of the cameras are "perfect." I don't expect any camera to be "perfect." But there's certainly something "a whole lot more wrong" with the S90 linearity than that of the other two cameras.

I really see what you could call "non-symmetrical non-linearity" in the S90 images, especially in the upper half of the images. I would call it, in the horizontal plane, in the upper half of the S90 images, it's like there is "pincushion distortion" on the left side (corner bending up), and "barrel distortion" on the right side (corner bending down). If my S90 had had "normal barrel distortion" on both sides, I'd probably just call it "more lens distortion than I'd like to see."

In the end, "whatever it is," it was way too much "geometrically wrong" than I'm willing to put up with. And I found it to be "easily noticeable," in comparison to all of the other camera I've ever owned.

Tom Hoots
http://thoots.zenfolio.com

Click to expand...

--

In Manchester in the 1840s, men and women were treated like animals. Why then should we be surprised that the utopian dreams of early communists were so appealing, or be so certain that they never will be again? -- Michael Elliott

Hi Tom,

Maybe there's an easy explanation for the odd geometric stuff going on in your S90 shots. If the sensor is tilted (and I contend that it quite obviously was) the distortion correction wouldn't know that, and still be applied assuming the file it was working on was level instead of slightly off. The result would be straightened edges that weren't perpendicular to begin with... and funky results on top of a slightly tilted image. How's that for a theory? I read of folks having to return/replace DSLRS from every brand due to tilted sensors. Actually, I'm surprised we don't hear of this more in small P&S cams, but DSLR users would be more attuned and critical of such things (plus, they're easier to hold level... we need Pentax's horizon leveler thingy from the K-7!) That was obviously a defective cam. It happens. If you try a new one, you might go to a different store and hope for a completely different batch.

Best regards,
Mark

--