Stitching with LCC: how to avoid differing color cast?

[rupho]

How do you guys get acceptable results when correcting 2 stitched images after applying LCC in C1?
I have 2 images one shifted upwards the other centered (unshifted)
Both images are exposed and processed the same same but when I apply LCC (to both images with their own respective LCC) I get not identical tonal values.

The shifted images where more color cast lifting has been done is lacking vibrancy compared to the images that was not shifted and as a result received less LCC color cast compensation in C1.

Is there a smart way to get both images corrected for color cast and falls of in C1 and still be able to stitch them in PS later on?

thanks for any input

Grischa

 

[Graham Mitchell]

Just curious - can you stitch the files seamlessly before correction? If you just shifted the back, this might be possible. Then you can manually correct the image as a whole.

 

[alan_w_george]

In C1...
Apply LCC correction
Basic adjustments (WB, curves, etc.)
Export TIFF

In photoshop....

File->Scripts->Load File into Stack....
Select all Layers
Edit->Auto-Align Layers (Reposition)
Edit->Auto-Blend Layers (Seamless Tones and Colors)
Merge all Layers
Crop a bit
Maybe some editing/spot removal/etc..
Done!

 

[Christopher]

or not.

 

[Jack]

In C1...
Apply LCC correction
Basic adjustments (WB, curves, etc.)
Export TIFF

In photoshop....

File->Scripts->Load File into Stack....
Select all Layers
Edit->Auto-Align Layers (Reposition)
Edit->Auto-Blend Layers (Seamless Tones and Colors)
Merge all Layers
Crop a bit
Maybe some editing/spot removal/etc..
Done!

Exactly, or use a dedicated stitching program like AutoPano -- regardless, it works like a charm, even with the problematic combo of the 43 SK lens at full shifts on the IQ180.

 

[rupho]

I am with Christopher On this one.
Under scrutiny it does not match exactly regardless of the stitching application.
And yes Jack you guessed it right the Schneider when shifted fully let's say around 18 mm can still be corrected but the LCC does need a lot of work to do which result in 2 very different looking images .

So it seems There is no elegant solution to accomplish 2 identically processed files without color shifts in C1 once the respective LCC has been applied?

 

[Jack]

No stitching matches "exactly" under pixel-level scrutiny. However for the most part they are darn close, especially if you employ one of the color and/or exposure blending routines in the stitching software.

 

[jlm]

so why not simply stitch pans, not shfts

 

[rupho]

Thanks Jack,
I attached my shot with the 2 different color read outs after LCC has been applied.
The subject matter (being a building) makes the difference all too obvious IMHO compared to a landscape shot.
here is the screen shot in C1:
Grischa

No stitching matches "exactly" under pixel-level scrutiny. However for the most part they are darn close, especially if you employ one of the color and/or exposure blending routines in the stitching software.

JLM: what do you mean by that:
"so why not simply stitch pans, not shfts"

 

[Christopher]

Grischa

I think you should add more image information about both shots. Otherwise some people might not understand what they are, or what you want to show. (I do)

 

[rupho]

Here we go:
The left shot is the upper half and right shot the lower part of a tall building
I picked a smaller crop of those images so the color read out is more visible
The left shot has been shifted up by 18 mm and I applied LCC and the right shot is not shifted and still got LCC apllied which resulted in different looking images. Yes both of them were exposed and processed the same way except for their respective LCC as mentioned earlier
jack send through a stitched landscape shot and while I can't see the any color cast I see the color cast on my image even after running it through various stitching protocols
I guess the nature of this subject matter with even colored broad areas such as the sky and the actual building itself it becomes more obvious
It also does not help to shift the 43XL on an IQ180 in that regard

I am just wondering if anybody here got similar problems

My temporary solution so far was to keep both images next to each other in C1 create color read outs at the exact same place and adjust the color and luminance channels on the other image to match my preferred stitch twin

But not very elegant to say the least

Grischa

Grischa

I think you should add more image information about both shots. Otherwise some people might not understand what they are, or what you want to show. (I do)

 

[jlm]

For the most part, I do pan movements of the entire camera to create overlapping shots, always using the center of the image circle, thus avoiding color casts and reduced optical performance near the edges of the image circle. Most easily done panning horizontally back in portrait mode, but can be combined with rise/fall to get vertical coverage

 

[gogopix]

Yes, that's the practical appraoch, and I find works well.

There are two reasons for shift for panning;
1. The shift (and I mean FP shift, not lens; lens shift causes parallax) with tripod is more stable (but with 80MP I am not sure anymore) and the stich can be more accurate That said, with panning PTGui (using pano tools) is still the most mathematically correct and does a great blend job. Pano factory and auto pan as well.
2. The second is an aesthetic argument that I have never seen tested; with shifts, you are creating a scene with a super wide, equivalent single lens (like the hassey wides) Panning, even thru noday point, greats a FOV akin to turning your head. I have never seen a convincing comparison to say they are really that different, or that it matters (but I am open.

What Jim has said, even handheld and vertical has produced great images (that SELL! ) because they are really wide, and big, so you have to move your head to see them. (like 2x6 feet)

regards
Victor

 

[alan_w_george]

Here we go:
The left shot is the upper half and right shot the lower part of a tall building
I picked a smaller crop of those images so the color read out is more visible
The left shot has been shifted up by 18 mm and I applied LCC and the right shot is not shifted and still got LCC apllied which resulted in different looking images. Yes both of them were exposed and processed the same way except for their respective LCC as mentioned earlier
jack send through a stitched landscape shot and while I can't see the any color cast I see the color cast on my image even after running it through various stitching protocols
I guess the nature of this subject matter with even colored broad areas such as the sky and the actual building itself it becomes more obvious
It also does not help to shift the 43XL on an IQ180 in that regard

I am just wondering if anybody here got similar problems

My temporary solution so far was to keep both images next to each other in C1 create color read outs at the exact same place and adjust the color and luminance channels on the other image to match my preferred stitch twin

But not very elegant to say the least

Grischa

In C1...
Apply LCC correction
Basic adjustments (WB, curves, etc.)
Export TIFF

In photoshop....

File->Scripts->Load File into Stack....
Select all Layers
Edit->Auto-Align Layers (Reposition)
!!!Edit->Auto-Blend Layers (Seamless Tones and Colors)!!!
Merge all Layers
Crop a bit
Maybe some editing/spot removal/etc..
Done!

Nothing is "perfect". You are using the best method and equipment for stitching (IMO). Don't throw the baby out with the bath water.

 

[David Kaufman]

I do all shooting with stitching using an Arca M-Line Two and 40+ back. Same problem. Applying LCCs even with symmetrical shifts off centre poroduces slightly different tonality. Even worse, with the Arca Swiss M-Line Two, the images where they overlap are of slightly different sizes (an NO slidinng back is being used, only the camera), by about 4 to 10 pixels maximum usually in one direction only, plus off slightly in orientation. I apply LCCs, make basic corrections, export to Photoshop and then resize one image manually and fix its orientation to perfectly align the two pieces. This leads to slight, very slight image degradation. Furthermore, I have tried using Photoshop CS3 to resize and align the images automatically but that version of Photoshop cannot do the job with most images. I then must do tone and colour matching manually. Sometimes Photosho will do it but not necessarily for all parts of the image, especially the parts that don't overlap. Final step in combining stitched images is distortion correction with Alpa's program, using custom text files based on final image size. Then you can refine the image. It's a lot of computer work, way too much work, but necessary to achieve the highest quality.

 

[Jack]

Another not-so-great image, but serves as an example (the weather is still rather un-cooperative). I assembled this 2-frame IQ180 stitch on my MacBook Air. It is 2 frames from my Arca RM3D with the 70 HR-W mounted, a 12.5mm right and a 12.5mm left. Exposure was 1/30th @ f8.3 for both frames, LCC was 1/15th @ F8.3 for both frames. Exposures were taken as quickly as possible, then LCC's were taken immediately after. The files were adjusted in C1 to the same settings, then LCC's analyzed, then LCC's added to the appropriate images, then the 2 files processed out. The tiffs were opened in CS5 and run through Photomerge using the "Collage" option for flat assemblies.

Note that dropping the two frames before assembly in the same areas generated extremely close readings, but not the exact same numbers, so there are some potential issues. However Photomerge does autocorrect for this during the actual assembly, and after the blend step I could not detect variance along the stitch-line. IMHO quite excellent for my needs, though YMMV for yours.

Here is the final at 1200px (about a 80th of the original!) for reference, stick line angles at a slight diagonal about 1/3 of the frame in from the left. Be curious to hear your thoughts. Oh, some more FWIW info -- final file is roughly 15000x7500 px and the pixel-level detail is astonishing:

 

[gebseng]

In photoshop....

File->Scripts->Load File into Stack....
Select all Layers
Edit->Auto-Align Layers (Reposition)
!!!Edit->Auto-Blend Layers (Seamless Tones and Colors)!!!
Merge all Layers
Crop a bit
Maybe some editing/spot removal/etc..
Done!

wow, this is super helpful, to say the least, thanks alan! works much better and is much more controllable than ye olde photomerge...

 

[Jack]

Those are the steps Photomerge does automatically

 

[gebseng]

I know, but (at least on my mac and with CS5), photomerge is sometimes not able to properly create a panoramic stitch. I often find myself trying the same action several times, and sometimes even restarting photoshop, before I get a result.
Also, the manual way as described allows me to manually adjust the parts to be stitched in density and color balance before blending, which is not so easy otherwise, and gives me more even results.

geb

 

[Jack]

Understood and agreed -- but honestly Photomerge has never had an issue doing a perfect job from a 2 or 3 frame shift-stitch for me