ColSebastianMoran
( IRL Richard Karash )
We may stitch for increased resolution, sometimes for 35mm but especially with 120 or larger format films. Sometimes we get distortions. Those who cam-scan with film edged report wavy lines. We have all seen distortion at times, subtle shifts.
I've been exploring distortion in stitching, and I think I now have a clean test result, and it's surprising.
Conclusion: For stitching of our camera scans, best results with a good lens of minimal distortion, a lens that does NOT communicate focal length to the body, stitch in LR (Perspective) or PS (Auto). Film curvature can create similar distortion; film flatness becomes even more important when stitching. Distortion from lens or film un-flatness likely to be magnified in the stitch.
My test: started with a good digital image, then:
- Add criss-cross lines
- Slice it into four parts by cropping, stitch these back together
- Add barrel distortion to each of the four parts, stitch these
- Add pincushion distortion to each of the four parts, ditto
- Then instead of cropping, do screen grabs to make the slices.
First, the original image:
Now the stitch from four screen grabs, these have NO lens info in EXIF, comes out perfectly to my eye:
Next, the stitch in LR of the four CROPs... It's distorted. Therefore, we can conclude that LR stitch is affected by the lens info (original was shot with 24mm lens, this lens info is passed into the crops):
Now the deliberately distorted crops, first from crops with a good dose of pincushion added, then stitched:
Finally, from the same crops with a good dose of barrel distortion added, then stitched:
I've been exploring distortion in stitching, and I think I now have a clean test result, and it's surprising.
Conclusion: For stitching of our camera scans, best results with a good lens of minimal distortion, a lens that does NOT communicate focal length to the body, stitch in LR (Perspective) or PS (Auto). Film curvature can create similar distortion; film flatness becomes even more important when stitching. Distortion from lens or film un-flatness likely to be magnified in the stitch.
My test: started with a good digital image, then:
- Add criss-cross lines
- Slice it into four parts by cropping, stitch these back together
- Add barrel distortion to each of the four parts, stitch these
- Add pincushion distortion to each of the four parts, ditto
- Then instead of cropping, do screen grabs to make the slices.
First, the original image:
Now the stitch from four screen grabs, these have NO lens info in EXIF, comes out perfectly to my eye:
Next, the stitch in LR of the four CROPs... It's distorted. Therefore, we can conclude that LR stitch is affected by the lens info (original was shot with 24mm lens, this lens info is passed into the crops):
Now the deliberately distorted crops, first from crops with a good dose of pincushion added, then stitched:
Finally, from the same crops with a good dose of barrel distortion added, then stitched:
charjohncarter
Veteran
I've heard you before talk about stitching, and I didn't get it. Now I'm even more confused. Why do you do it in the first place? Years ago (2008) when I 'cam-scanned' I made a holder for my 120 film and shot it with a macro lens. I remember really being careful with my squaring of the holder which wasn't attached like my 35mm slide coper was.
shawn
Veteran
You stitch to get more resolution. If I shoot a 6x9 negative with one frame on my Sony A7RII it is a 42ish megapixel file. If I shoot the negative at 1:1 ratio and stitch if will be 4-5x that size. Not always worth the hassles but sometimes it can be.
Shawn
shawn
Veteran
Nice write up. I've also found that making sure the stitching lens isn't vignetting at all is also really important when stitching a larger negative. Otherwise you can end up with obvious problems in skies and such. Important for single shot digitizing too if you are trying to fill the sensor with the negative. For the 90mm Tamron macro lens I use I need to be at f11 to avoid this.
Shawn
ColSebastianMoran
( IRL Richard Karash )
Fair question. Frankly, at this point is for curiosity, what conditions create the distortion, how to beat it.
For 35mm, my 24MPx cam-scans are all I need.
And, even for 120, I have a hard time seeing a different between 24MPx capture and a 100MPx stitch printed at 20x30". So, yeah... it's mostly my curiosity, and I'll stitch a few selected 120 images.
charjohncarter
Veteran
I don't use my FF digital camera for 'cam-scan' but I wonder if pixel shift would give you what you are doing?
shawn
Veteran
Pixel shift would give more color information but depending upon how it is implemented may or may not increase spatial resolution. Some systems do, some don't. You could of course also combine pixel shift images with stitching too.
The other situation stitching can be beneficial is when the negative is a different AR than the AR of the digital camera. For example when I shoot 35mm panoramics in my GSW690 the AR is close to 3.5:1. If I capture the negative in a single shot I'm throwing away more than half of the vertical resolution of the sensor so the resulting image ends up at less than half the resolution of the digital camera. But if I instead fill the sensor vertically and then stitch horizontally across the negative I end up with an image 2 or 3x the resolution of the camera.
Shawn
The other situation stitching can be beneficial is when the negative is a different AR than the AR of the digital camera. For example when I shoot 35mm panoramics in my GSW690 the AR is close to 3.5:1. If I capture the negative in a single shot I'm throwing away more than half of the vertical resolution of the sensor so the resulting image ends up at less than half the resolution of the digital camera. But if I instead fill the sensor vertically and then stitch horizontally across the negative I end up with an image 2 or 3x the resolution of the camera.
Shawn
charjohncarter
Veteran
Is spatial resolution MP count? And is that important in only very large finished prints? Sorry, for being so ignorant on this subject but I'm interested even though I'll never do it.
shawn
Veteran
Spatial resolution is the overall x number of pixels by y numbers of pixels. Pixel shift doesn't always increase that. In some systems what it does is get rid of the Bayer interpolation of RGB by sampling each of those colors at every pixel location. Some pixel shift systems do that and also sample RGB at a slightly different position to increase overall resolution. To do this of course takes multiple photos over a period of time so for a dynamic subject it might not work as well. But for scanning negatives that wouldn't be an issue.
Shawn
charjohncarter
Veteran
Thanks, mine takes four photos and then I get a huge DNG. I then post process with RAWtherapee using their pixel shift module ending up with a TIFF that is even bigger. In the end I have to resize it to use it.
It seems like over kill for a 4x6 or 5x7 final photo. But I will say the resolution is really good.
Full size pixel shift DNG (converted to JPEG for the web):
Pentax 35mm f2.0 FA-AL by John Carter, on Flickr
Small crop from the pixel shift image:
Pentax 35mm f2.0 FA-AL by John Carter, on Flickr
This was an f2.0 shot using a 35mm (f2.0) Pentax lens.
It seems like over kill for a 4x6 or 5x7 final photo. But I will say the resolution is really good.
Full size pixel shift DNG (converted to JPEG for the web):
Pentax 35mm f2.0 FA-AL by John Carter, on FlickrSmall crop from the pixel shift image:
Pentax 35mm f2.0 FA-AL by John Carter, on FlickrThis was an f2.0 shot using a 35mm (f2.0) Pentax lens.
shawn
Veteran
The Pentax system is increasing color information (each color captured at every pixel location) but not increasing overall spatial resolution as it doesn't go back and do the same thing again at another slightly differing 'pixel' location. This explains the difference pretty well.
https://www.imaging-resource.com/PRODS/pentax-k1/pentax-k1PSR_MODE.HTM
The sharpness of all colors at each pixel location is why Sigma Merrill cameras have their followers in spite of the other limitations of the cameras.
Shawn
https://www.imaging-resource.com/PRODS/pentax-k1/pentax-k1PSR_MODE.HTM
The sharpness of all colors at each pixel location is why Sigma Merrill cameras have their followers in spite of the other limitations of the cameras.
Shawn
charjohncarter
Veteran
Thanks Shawn.
charjohncarter
Veteran
I just took my sixth pixel shift image, and I've had the camera for 2 years.
DNG 157MB
Pixels 36.3 MP (I think: spatial resolution 36,300,000; 7372x4924)
TIFF 208MB
JPEG 32MB
Seems like a lot to me more than I would ever want to deal with. What do you guys use for a 20x30 inch print? I would guess 54,000,000 pixels or 54 MP: is that right?
DNG 157MB
Pixels 36.3 MP (I think: spatial resolution 36,300,000; 7372x4924)
TIFF 208MB
JPEG 32MB
Seems like a lot to me more than I would ever want to deal with. What do you guys use for a 20x30 inch print? I would guess 54,000,000 pixels or 54 MP: is that right?
shawn
Veteran
Pixel shift isn't just for printing big though. If you have an image that has fine detail that would produce color moire artifacts the pixel shift image *should* eliminate that.
Shawn
Shawn
charjohncarter
Veteran
I know, so I wonder what is it for (showing off)? Like I said I have only used it six times in two years, for nothing other than to see if it works (which it does). I didn't even know what pixel shift was when I bought the camera and it wasn't a selling point for me.
I guess I could use it for 'cam-scan' but I would have to retro fit my set up.
I guess I could use it for 'cam-scan' but I would have to retro fit my set up.
ColSebastianMoran
( IRL Richard Karash )
I'm getting very satisfactory 20x30" prints from 24MPx both original digital captures and camera scans. I made a recent post: 24MPx one shot vs 100MPx stitch from 6x9 are very hard to tell apart in 20x30" print.
Stitching and pixel-shift for me is only for a very few images and maybe for experimentation.
charjohncarter
Veteran
Thanks you Richard. I doubt that I will every use Pixel Shift for a finished print. But if I do I'll PM you for advice.