How to Fix Sensor Tiling With Photoshop

[diggles]

My first choice for photo editing is ACR/Photoshop. While looking for a way to apply LCC corrections without using C1/Phocus/Lightroom I ran across this article:

Vignetting 4: Photoshop Divide Layer

Vignetting 1: Overview Vignetting 2: Parametric Correction Vignetting 3: Correction Images Vignetting 4: Photoshop Divide Layer (this post) Vignetting 5: A Pathological Example Vignetting 6: Other …

pduncan.com

The steps outlined in the article do a good job of fixing vignetting, but didn't really help with colorcast. At first I was a bit frustrated, but then I wondered if this method could work to fix sensor tiling. I've tried it on 4 different images, 1 taken with the Digaron 28HR and 3 taken with the Digitar 43XL. Initial tests are quite promising.

Here are the steps I took with example images and screenshots:

STEP 1
RAW CAPTURE: Hasselblad CFV II 50C (cmos 50mp sensor), Digitar 43XL, and the WRS 1600 with 20mm of camera fall

RAW CAPTURE 100% Crop

As you can see this combo creates one hell of a mess. Tiling is very pronounced and streaky noise is noticeable in the sky.

Matching LCC Frame


STEP 2
Process files in Phocus and export 16bit tiffs to edit in Photoshop. With Phocus you are able to apply the LCC correction to the LCC reference file and export the corrected image as a 16bit tiff. This step is critical in making this work. Not sure it would work without having a corrected LCC reference file to work with. I'm not sure if Capture One or Lightroom allows you to export the corrected LCC reference file or not.

Corrected image using Scene Calibration in Phocus exported as 16bit tiff file


Corrected image 100% crop


Corrected LCC reference image


Phocus did a pretty good job of cleaning the image up. The extreme color cast is gone, but the sensor tiling and streaky noise is still present.

STEP 3
Create a photoshop file with the corrected image as the base layer. Add the corrected LCC reference image in a group above. Then create a Levels adjustment layer above the corrected LCC reference image. In the Levels control, hold down the option (alt on a PC) key and move the highlight (white) triangle slider left until you start to see highlights being clipped. The highlight triangle will end up being just to the right of the histogram mass.

Screenshot of the layer configuration and histogram


After making the levels adjustment, this is what the LCC reference image should look like


Step 4
Change the blending mode of the group with the LCC reference image and Levels adjustment layer to Divide and fine tune if needed.

Screenshot of blending mode changed to Divide


That's it. I'll add the final images + 100% crops in a follow up comment.

Note: With this image I did not need to fine tune the settings. With a different image I tested, there was still a faint outline of the tiling present after changing the blending mode to Divide. What I did to fix it is move the shadow (black) triangle in the levels control to the right until the tiling disappeared.

 

[diggles]

Here are the final images with 100% crops.

After applying the levels adjustment to the corrected LCC file and change the blend mode to Divide


100% Crop


After making final corrections to the image in Photoshop


100% crop


The tiling and streaks are gone. There is still noise in the sky, but there are other ways to fix that.

I'm not sure if this will work on any image, but so far it has worked for me on 4 different scenes. This example is the most extreme that I tried it on. If you have any images with tiling, try to see if this fixes it.

 

[dchew]

Warren,
I am so going to try this with a few 35xl images. Thank you very much for posting!

BTW, you can export a corrected LCC in C1 just like a normal image.

Ciao,
Dave

 

[diggles]

Good to know that you can export a corrected LCC in C1, I assumed so, but didn't know for sure.

Hope it works with the 35xl too. Post an example when you get a chance

 

[usm]

This sounds interesting. Will try it also with the XL35. Thanx

 

[usm]

Warren,
I did a quick test with my CFVii50c and the Schneider XL35 - it works with the sensor tiling problem - really great - thank you!!

But it doesn't help with the pattern problem. Did you use the "Remove Checkerboard Patterns" in the "noise filter" tool?
I can see red colorshift on your untouched image in top left corner. Is this just a color shift or are the strange patterns?
Like this:



Mario

 

[kinglang]

"Remove Checkerboard Patterns" will work fine

 

[diggles]

I did a quick test with my CFVii50c and the Schneider XL35 - it works with the sensor tiling problem - really great - thank you!!

But it doesn't help with the pattern problem. Did you use the "Remove Checkerboard Patterns" in the "noise filter" tool?

Hey Mario, I'm happy to hear the sensor tiling fix worked for you, great news! I didn't use the "remove checkerboard patterns" option. The pattern shown in your example wasn't present in my image. I have seen that pattern in images before, but didn't realize that there is a checkbox in Phocus to mitigate it. Good to know.

I'm curious under which situations you've ran across this pattern? I ran across this pattern while doing some tests when I first got the lens. For me, it was when shifting left and right 12mm or more while the camera was in vertical position. I deleted those files, but I'm going to try and recreate it so I can see how well the "Remove Checkerboard Patterns" works for me.

"Remove Checkerboard Patterns" will work fine

Hey Kinglang, do you have a "Remove Checkerboard Patterns" before and after example you would be up for sharing?

 

[usm]

The "remove checkerboard patterns" option works for me, but it is strange:
Movement is always 15mm which is the max on my factum:

Shifted up and down, Sensor horizontal - I have to turn it OFF! When turning ON the pattern is present.
Shifted up and down, Sensor vertical - I have to turn it ON!
Shifted left and right, Sensor vertical - I have to turn it ON! But the patter is NOT removed completely.
Shifted left and right, Sensor horizontal - I have to turn it ON! Pattern is removed.

Shifted right, Sensor vertical and shifted up, Sensor horizontal is the same part of the lens and the same part of the sensor.
For me, it is a bug in Phocus.

 

[diggles]

That is strange. Do you know if the checkerboard patterns only show up on Hasselblad files? Or do they show up on other camera files that use the same 50mp CMOS sensor?

 

[Paul Spinnler]

Thanks Warren, will check this out! The 43XL and the 60XL are the lenses which allow you to do massive stitching ... so very helpful indeed

 

[kinglang]

an example, shot with HR70 & Hassy back,It is more likely to appear when the moving range is greater than 15mm.Checkerboard patterns are also sometimes visible from back screen playback when using P1 back, but they disappear when imported into C1. So I think C1 automatically eliminates them by default.

 

[diggles]

Thank you for sharing this Kinglang. I wasn't expecting it to be an issue with the HR and 70mm focal length. I was assuming it was a side effect of the Schneider wide angle lenses.

 

[usm]

©diggles:
I tried your solution with Affinity but it doesn't work.

The Level correction on the LCC image works with Divide blend mode.
But when I turn on the background image it does this:
Image:

LCC screenshot:


Affinity LCC Divide:


Affinity LCC oder Image in blend mode Divide:



Any ideas regarding Affinity?
Thanks. Mario

 

[diggles]

Hey Mario,

Unfortunately I'm not familiar with Affinity.

Have you tried Marco's fix?

Hasselblad 100C and 35XL

Based on playing around with the files and looking at the histograms of rows, I think they probably could write an algorithm that cleans up the shadows. Basically detect when a set of rows has a big change in average brightness. Should work unless you photograph very regular horizontal lines :)...

www.getdpi.com

 

[usm]

No I didn’t. But do think this will also work for sensor tiling? Will try…

 

[mristuccia]

No I didn’t. But do think this will also work for sensor tiling? Will try…

Well, honestly my solution was specifically tailored to fix the PDAF banding.

The way this solution works is the following:
it should be able to re-gain the original brightness of whatever pixel is darker or brighter than it should be. Anomalous darker and brighter pixels are clearly identifiable in the LCC, once it is self-applied and white balanced so that the average areas with no issues are exactly 50% grey. Since such LCC should be perfectly uniform, all the anomalies will be brighter or darker, and the inverted LCC in overlay mode on top of the original image will counter-compensate them, effectively simulating a very localized dodge and burn, and keeping the rest of the image (where corresponding LCC pixels are exactly 50% grey) unchanged.
A contrast clipping layer applied (only!) to the inverted LCC will keep the 50% grey as it is while exaggerating darker and brighter pixels. This allows to fine tune the counter-balance effect.

Now, if sensor tiling results in darker or brighter pixels on the otherwise uniform LCC image, this solution could work. In this case the only challenge I see is choosing the correct brightness of the LCC image. I think you can try targeting 50% grey on one of the tiled squares of the chessboard, possibly one of the darker ones. All the rest of the LCC pixels will move accordingly. Then invert the LCC image and proceed applying my solution.

Let me know.

 

[usm]

Tested both variants and prefer the version from Marco. With the divide version from diggles applied two times it changes the color of the sky.
Thanks both for the work!

no correction:

LCC divide two times:

LCC inverted overlay contrast 95: