Pixel Shift with the D850..... and 85mm PC?

Just checking the theory here before trying a somewhat elaborate experiment.....;)

 

If (!!!) I can securely clamp my 85mm PC to a heavy tripod head, and leave the D850 free, can I take a frame at the limit of shift in all directions (and the central frame) and stitch them together to form something like a 100MP single image?

 

This would be for completely stationary subjects such as fine art or still life.

 

I will have to provide some support to the rear half of the lens plus the camera body combo as the extra strain on the tilt/shift mechanism will be dangerously high.

If i understand your description, th front lens of your lens will not move at all (theoretically) and the camera will rotate with thr rear part of your lens?

Then all pics will be almost overlapping fully with the same part of the subject ?

 

anyway you can greatly reduce the forces on the lens by having the body of your camera rest on a table of sorts or having it attached by a trypod on a dolly i guess.

Thom Hogan has written about a product that lets you do this sort of thing. Stitching Z Simple Way | Thom Hogan

Shifting a few microns is an impossible task without the benefit of IBIS, or a similar mechanism (e.g., Phase One). Sometimes we have to bite the bullet, so to speak, and buy equipment which does what we need (or want) it to do. It took me 35 years to abandon Leica in favor of the flexibility and precise framing of an SLR. 15 years later the SLR gave way to mirrorless, and each subsequent upgrade has been without pangs of nostalgia.

Shifting a few microns

Maybe I should have called it Super-Sensor Shift, in as much as the sensor's movement in relation to the optical lens axis is ~ +/- 12mm from normal.

 

The aim is to take full advantage of the much larger image circle without changing optical perspective.... resulting in a single image with a much higher pixel count.

 

If you follow the link provided by NetR above, it says it much better!

Unusually, I didn't find Thom's explanation very clear about the amount of shift used and what coverage one could expect and maintain quality. The method also relies on lens performance outside the optimum image circle, I think. I've never had much downside when doing ordinary panoramas by swinging the whole camera. Modern software has no trouble with this most of the time.

 

I have an upcoming shoot for a trade show booth photo that needs high resolution. Ideally I'd have a 60 MP camera, but don't. I have my Z6 at 24 MP. I'll probably stitch two or more images, but if there's a live subject, that won't be possible. In that case I'll do the best I can and use good interpolation software. The end result will be fine. The problem is the expectation is to view a huge image from a foot away, IMO, unreasonable. What I wonder about, in real life, is can you even put a good enough image on a 60 MP sensor? The subject isn't going to be flat, so DOF comes into play. Lenses are good today, but I probably won't be at the optimum aperture. There may be live/moving subjects. There's never enough light so the ISO won't be optimum and we'll have a bit of noise. Bottom line is you always have to compromise and getting pixel level sharpness at 60 MP seems like a pipe dream.

Maybe I should have called it Super-Sensor Shift, in as much as the sensor's movement in relation to the optical lens axis is ~ +/- 12mm from normal.

It's much simpler if you move the lens with the camera stationary, and does exactly the same thing. Take four images and do a 2D stitch (e.g., PtGUI).

 

 

Alternately, borrow or buy a Sony A7Riv and get a 270 pixel image using Pixel-Shift.

Technically, it doesn't do the exact same thing, but in any normal application I don't think the difference amounts to even a small hill of beans.

OK, I'll do the experiment both ways and report back!

 

I suspect there may be issues with the different perspective (with lens shift) with the depth of a picture frame or flowers 'parralaxing' in a vase.

It will be interesting to see (or hear about) the results. My guess would be that you will gain MP, but lose actual resolution due to tiny lens movements while shifting, using the less than optimal portion of the lens image circle, and other factors that rob potential resolution.

Technically, it doesn't do the exact same thing, but in any normal application I don't think the difference amounts to even a small hill of beans.

True, but if the closest object is 15' or more away, the parallax would be nearly unnoticeable.

 

While it might not be feasible to hold the lens steady (in lieu of a tripod collar), you could use a 3-way pano head to balance the camera and shift the camera vertically and horizontally to offset movement of the objective.

 

Full Size Pano-Gimbal Head

Indeed, at 15' parallax is minimal, but I'm probably at 6' for the fine art and maybe 4' for the still lives.

 

I found a manufacturer that makes an Arca Swiss footed holding bracket that clamps on the lens body and allows movements.

 

They make similar brackets for other PC lenses. Look up iShoot PC-85 for this one.

mm still trying to imagine the advantage compared by just using a bit wider lens...and moving the whole camera and lens a bit to the left and the right and stitch the pics .., would need to see both results in real before i can immagine the result ...

still trying to imagine the advantage compared by just using a bit wider lens...and moving the whole camera and lens a bit to the left and the right and stitch the pics

Theoretically, the answer is parallax from the differing viewpoints recording as ghosts and very difficult to align..

 

2D or flat planes should be fine, but I don't know how much it will effect 3D objects, especially sparsely spaced things like flowers in a vase.

It will be interesting to see (or hear about) the results. My guess would be that you will gain MP, but lose actual resolution due to tiny lens movements while shifting, using the less than optimal portion of the lens image circle, and other factors that rob potential resolution.

The premise of this thread has evolved from pixel-shifting (in increments of a few microns) to image shifting to make full use of the large image circle of PC lenses. There is a PC adapter for Hasselblad V lenses, which have a comparable image circle nearly an inch larger than a native lens (+/- 12 mm). Shooting at the extreme limits and stitching the results would yield more usable pixels, but only about 2x for 4 frames, due to considerable overlap.

 

This process would not require more resolution from the lens, since there is no magnification. whereas 16 frame in-camera pixel shifting would require very resolution from the lens to take advantage of the difference, which includes elimination of color aliasing.

 

Since the lens is not rotating, there is no need for a nodal plate, nor cylindrical or spherical correction.

Edited August 15, 2021 by Ed_Ingold

Technically, it doesn't do the exact same thing, but in any normal application I don't think the difference amounts to even a small hill of beans.

Depends how close the subject is. Because there's no difference between moving the optical axis by some distance, and moving the subject the same amount. So if you're shifting the lens by 12mm, that's the same as the subject being shifted by 12mm relative to the camera. And that's not making any allowance for lens distortion disturbing the match between shifted images.

 

The technically correct way is, as Mike originally proposed, to hold the lens still and move the sensor around the image plane by using the shift mechanism.

 

There were shifting back gizmos made for LF cameras for this exact purpose. The idea being that you bayonetted your DSLR to the back of your 5x4 Sinar, or whatever, and got a high megapixel-count stitched image. With the limitation that the sensor was going to be vignetted by the digital camera's lens-throat at some point - which the pushers of such systems neglected to mention!

 

I see no reason why using a lighter and smaller-sensor body wouldn't work just as well as the D850. It's the pixel density that counts; although you might have more frames to stitch.

Edited August 15, 2021 by rodeo_joe|1

smaller-sensor body

That's a lot of frames with my J5!

The premise of this thread has evolved from pixel-shifting

Sorry, that was only my badly worded title.

 

I certainly never intended any sub micron movement!

The problem is the expectation is to view a huge image from a foot away, IMO, unreasonable

The issue I have is a client that wants to project (4K) and allow users to zoom in to see even more detail.... Think continually zooming into the Mona Lisa from full size down to the micro brush strokes..... seamlessly anywhere on the painting.

 

The display of images is no longer a static thing, such as a single big print, it's a dynamically scalable entity.

 

I may have to go MF!

I think I need to go review the diffraction formulas!

The gain in pixel count through lens shifting is relatively modest. What is the largest 3:2 image you can fit in the image circle? The image circle for a 24x36 mm image is at least 49.5 mm (the diagonal). The PC lens has an estimated image circle of 66.5 mm.66.5/49.5, which yields an estimated 1.8x gain in pixel count. Resolution of the lens is not a factor, since each component of the image is captured at the same magnification.

 

Diffraction is fairly easy to calculate. The Airey circle diameter is proportional to wavelength x the relative aperture. The constant of proportionality is 1.22. However you are not multiplying the number of pixels per sensor area, rather adding 4 normal images together with considerable overlap.

I've played with this a couple of times. I got an adapter that allows me to mount my old Bronica medium format lenses onto Nikon F mount bodies (I use the D850), and it allows a shift of up to 11mm I think, to make use of the larger image circle of the MF lens. It's fun, and I got nice big pictures, but practically I'm not sure I got more resolution than I would have with a sharp dedicated Nikon lens.

 

I did not bother to fix the lens in position and shift the sensor behind it, rather just put the camera on a tripod and let photoshop deal with the very insignificant parallax error I got from shifting the lens.

What distance were you working at?

 

Close range shows the affects of parallax the most. Landscapes will be fine.

 

I'm pretty sure parallax is going to mess up even the mighty Photoshop. How's it supposed to know which image is true?

Edited August 16, 2021 by mike_halliwell

If the axis of rotation is centered on the front node* of the lens, there is no parallax due to rotation. There are fixtures which allow the camera to be centered on two axes, so a 2D array of images can be gathered and stitched. Unlike with lens shifting, all of the images are centered, with no more vignetting and aberrations than for a single image. Whereas squeezing multiple (e.g., 4) exposures into an existing image circle, with 50% or more overlap, you need only 25% overlap for a good stitch.

 

* The front node is not always a single point and may vary with the angle of incidence. Consequently residual parallax may still occur with rotation.

If the axis of rotation is centered on the front node* of the lens, there is no parallax due to rotation.

It is.

 

That's why I'm intending to rigidly fix the lens and rotate/shift the camera body.

 

As I said at the start....;)