Stitching with a tilt-shift lens

[Kirk Gittings]

There is no feature of a tilt shift lens that cannot be replicated with current high quality bodies, lenses, software, and tripod heads.  If you want a TS lens as a way to inspire creativity, go for it. If you think you will create more perfectly sharp images with less fuss, think again.

As a practitioner (35 years) of and professor (25 years) teaching architectural photography at two universities I have both participated in the genre from film to digital and been in a position to watch and contribute to people developing their AP skills.

The problem, as I have experienced it, with doing everything like perspective correction in software is one of vision. From what I have experienced personally and with students, the more you can do in camera the stronger things like composition etc. will be. That is why largely, IMO and IME, all of the top APs that I know (and I know a ton worldwide) all use MF cameras with movements and/or DSLRs with T/S lenses.

[Ellis Vener]

Off topic slightly (and I agree with Kirk) but has anyone tried the Lightroom plug-in set "TiltShift Presets?  http://www.adobe.com/cfusion/exchange/index.cfm?event=extensionDetail&loc=en_us&extid=3265522 ?

[Scott Hargis]

There is no feature of a tilt shift lens that cannot be replicated with current high quality bodies, lenses, software, and tripod heads.  If you want a TS lens as a way to inspire creativity, go for it. If you think you will create more perfectly sharp images with less fuss, think again. 

You're assuming that everyone only shoots landscapes. For architectural work, you can't really software your way around a shift lens. Yes, there's perspective correction, but you can only stretch pixels so far. I guess it really comes down to the level of IQ you can live with. If I were to manipulate perspective on files from my 5DmII to replicate what I can achieve with a TS lens, I might end up with a file that's about as good as what I used to get with my classic 5D (about 12MP). Which no longer seems acceptable to me.

Someday we'll have sensors of such quality that we'll only need to own one lens, maybe a 10mm. Everything will be cropped and digitally manipulated from that!

Or maybe not. I sort of hope not.

[Petrus]

When I wrote that shift can be done in software (either stitching or perspective control by warping) I meant that identical projection can be done that way, but naturally the IQ will suffer if we start stretching a single frame. With stitching we might (will...) get better IQ than what we get from shift lens. When taking a picture all 3D information is lost and we can not exactly replicate the effect a tilt lens would have had because the captured projection would have been different with a tilted lens. With shift we are using the same projection but just extending it (or recalculating it when doing rectilinear stitch), but software manipulated fake tilt does not have the same projection parameters as a real tilt photo, as the lens plane is not the same and there is not 3D information in the picture to synthesize it perfectly.

[Ellis Vener]

When I wrote that shift can be done in software...

What you are referring to is changing the apparent convergence angle of parallel lines.

with a camera this is accomplished by changing the angular  relationship between  between  the planes of the subject and the film/sensor. Software does this by shifting data points and interpolation, either creating new data points between the existing ones or eliminating existing ones,   

What lens (front) or film/sensor (rear) shift does cannot be emulated by software as by physically moving  the lens or the film/sensor moves the lens axis from the center of the frame,   changing what part of the subject is  framed on the recording medium.

[John Rodriguez]

There is no feature of a tilt shift lens that cannot be replicated with current high quality bodies, lenses, software, and tripod heads.  If you want a TS lens as a way to inspire creativity, go for it. If you think you will create more perfectly sharp images with less fuss, think again. 

You can fake tilt DEfocus, but you can't fake tilt focus with a single shot.  You also can't fake shift in software with a single shot, and in some cases you can't at all (lenses only go so wide).  The best answer is to use whatever tool is necessary for the situation. If you don't have things changing in frame then a pano head + focus stitching makes sense.  If things are changing then a T/S is the way to go.  If you already own T/S lenses then you can do shift panos, with the trade-off of lack of "un-distorting processing" resolution loss vs "edge of lens coverage" resolution loss. I'd prefer to have T/S lenses AND a pano head.  If I could only have one I'd take the T/S lenses - at least you can always get the single shot.

T/S in landscape shots: Personally I use rise on at least 50% of my landscape shots (I shoot a lot of trees).  I use tilt less often, but often enough to know I need it.  After shooting field view cameras the geared DSLR T/S lenses are pretty quick/easy to use.

T/S adapters - I'm aware of Mirex, Harblei (not available anymore that I can tell except Ebay) and Kipon.  There are wide-angle limitations. From what I can tell all have the same issue as the PC-Es, shift + tilt OR rise/fall + swing.  Not a big deal to me, as I rarely really NEED to use rise and tilt or rise and shift or tilt and swing (never) at the same time, but an issue for some folks.

Metering T/S shots - metering seems fine using the Nikon PC-Es on a D800e tilted and shifted, however I often still use a dedicated spot meter. 

[Petrus]

What lens (front) or film/sensor (rear) shift does cannot be emulated by software as by physically moving  the lens or the film/sensor moves the lens axis from the center of the frame,   changing what part of the subject is  framed on the recording medium.

I'll try to go through this once more: Let's compare a single shifted shot taken with a TS lens or technical camera against a rectilinear stitch panorama and a perspective corrected single shot taken with a tilted camera.

It is the position, aim and focal length of the lens which determines the geometry of the picture (assuming no tilt). Taking several shots with a longer lens and stitching them as a rectilinear composite can be made identical to that first shift picture (I am talking only geometry, not IQ), if the position of the front nodal point of the lens is identical. That is important. Same thing with a tilted single shot which is corrected in software, as long as the angle is wide enough to have room for the correction. Out of these three methods the stitch will give the best IQ if we are using the same camera and the lenses have identical IQ.

If you mean that we first take a normal shot with T/S lens and do a stitch from the same position, they are identical. Then we shift the lens, take a shot and compare with the said stitch photo. Of course they are not identical anymore, as the position of the lens is not identical anymore. If we shift the back, not lens, they remain identical after framing is adjusted.

[Ellis Vener]

Petrus we are discussing different things.

"If you mean that we first take a normal shot with T/S lens and do a stitch from the same position, they are identical. Then we shift the lens, take a shot and compare with the said stitch photo. Of course they are not identical anymore, as the position of the lens is not identical anymore. If we shift the back, not lens, they remain identical after framing is adjusted."

No I meant what I said. Shifting a lens vertically, laterally or to some intermediate angle  changes the angle of view framed by the recording medium. Lenses with shift function project a relatively large diameter image circle - shifting the position of the lens or the recording medium changes the portion  of the image circle framed by the edges of the media.  Pointing the camera in a different direction has the same effect. This is simple geometry and has nothing all to do with stitching multiple frames together. Tilting the lens or the film/sensor plane can change the framing of the subject as well.

[NancyP]

My local camera store has or used to have a Canon TS lens for rent. If it is still around, I am now curious enough to give it a try. I have the feeling that no amount of blog perusing will help me see if TS is something I may want to use more often. 

[Ellis Vener]

My local camera store has or used to have a Canon TS lens for rent. If it is still around, I am now curious enough to give it a try. I have the feeling that no amount of blog perusing will help me see if TS is something I may want to use more often.  

"One test is worth a thousand internet thread postings by a bunch of opinionated old guys." (he said while looking at his reflection in a mirror   )

[Jim Kasson]

Someday we'll have sensors of such quality that we'll only need to own one lens, maybe a 10mm. Everything will be cropped and digitally manipulated from that!

Physics rears its ugly head, in the form of diffraction.

[Scott Hargis]

I always hated physics.

[Ellis Vener]

I always hated physics.

"Physics. Why does it always have to be physics?" 

[Jim Kasson]

"Physics. Why does it always have to be physics?" 

I was a Physics major at Stanford for two and a half years before I switched to Electrical Engineering. I don't like it any better than you do.

[pjtn]

The biggest issues to me seem to be parallax and lens edge performance. How close does a foreground subject really have to be before parallax becomes and issue? I would have thought the 50% overlap from the three shots would be enough to prevent this.

The corner performance there is obviously no way around. I can't test one of these lenses so I can't be sure how bad it really is.

On a different note, we have the ability to stitch to create higher megapixel images, is there a method to increase the tonality of a DSLR? I love the smoothness of tones medium format film and digital have.

[John Rodriguez]

The biggest issues to me seem to be parallax and lens edge performance. How close does a foreground subject really have to be before parallax becomes and issue? I would have thought the 50% overlap from the three shots would be enough to prevent this.

The corner performance there is obviously no way around. I can't test one of these lenses so I can't be sure how bad it really is.

On a different note, we have the ability to stitch to create higher megapixel images, is there a method to increase the tonality of a DSLR? I love the smoothness of tones medium format film and digital have.

To avoid parallax issues with shift panoramas you can use a nodal slide and move the camera back the same number of mms you shift the lens.  

As fars as increasing "tonality", what I think you're referring to are the levels of gradation the sensor can display.  Medium format cameras are usually 16 bit, while DSLRs are generally 12 or 14 bit. Stitching doesn't increase bit depth.  Film isn't binary.  If you want absolute maximum separation of tones and tilt shift lenses, buy a view camera and a freezer full of E6.

[Jim Kasson]

On a different note, we have the ability to stitch to create higher megapixel images, is there a method to increase the tonality of a DSLR? I love the smoothness of tones medium format film and digital have.

We can make the sensor bigger, but you want to keep the sensor size to 24x36mm, right? We could make the photosites bigger, but that's fewer pixels total. We can't do much about photon noise; that's more inconvenient physics. We could increase the exposure to get more photons, and thereby increase the signal-to-noise ratio (SNR) by a factor of 1.4 for every stop. But then we'll blow out the highlights. So maybe we can make a series of exposures a stop or two apart, and use the high SNR data from each to create a high SNR composite image. Oh, wait; that sounds familiar. They call it HDR. You can get some really unnatural images that way. But what if you used the technique just to improve the SNR on subjects of normal dynamic range?

OTOH, this could be one of those be careful what you wish for things. You've already said your exposures are uncomfortably long, and this will add longer ones. You'll have to decide, composite the exposures first and then stitch, or the other way round (I recommend the former). If you do focus bracketing, there will be a bunch more exposures. A lot to keep track during the shoot. A lot to keep track of after the shoot. A lot of opportunity for error. Hmm...

[Petrus]

On a different note, we have the ability to stitch to create higher megapixel images, is there a method to increase the tonality of a DSLR? I love the smoothness of tones medium format film and digital have.

As far as I know the best digital sensors (did somebody whisper "D800"?) far surpass the dynamic range of film, and there is no lack of smoothness.

[pjtn]

Jim, I would like to keep the sensor 24x36mm for price reasons. So you're saying maybe 3 half stop increments in exposure, then stacked might create a smoother tonality? I have seen the same sort of thing done to reduce noise in an image.

Petrus, personally I find the DSLR images to be crunchy and lack the three dimensionality of film images. Have a look at some of the photos in this collection by Hin Chau http://www.hinius.net/atf.php. I believe he was using a Mamiya 7 II. Or am I seeing something else?

Here's a link to one in particular http://blinding.slideshowpro.com/albums/001/390/album-122821/cache/Shanghai-18-4445-03.sjpg_800_640_0_90_0_50_50.sjpg?1360205836

[John Rodriguez]

As far as I know the best digital sensors (did somebody whisper "D800"?) far surpass the dynamic range of film, and there is no lack of smoothness.

I own a D800e and a 4x5.  I can get more range with Portra 160 or 400 then I can get with the D800.  The dynamic range numbers tossed around DSLRs is a bit of a misnomer.  Yes you can recover a lot of detail in the shadows shooting RAW, but you generally don't want to use all of it if your goal is a really good large print as there isn't much tonal separation down there.  I find the D800 closer to E6 in terms of usable range.  In terms of gradation of tones, E6 shows more tonal variation, there's no doubt.  In my opinion the D800 looks closer to C41 in that regard.