because the VueScan algorithm ist not well suited for Kodachrome. It also leads to a reduced contrast in the output, as you might see in your examples.
If you ask VueScan to display the defects found in the preview window, it only shows you the most prominent defects. I asked EH about this and he admitted, that the actually corrected pixels are more than those shown in the preview …
Hermann-Josef
Hah, I sent him an email asking exactly this question literally seconds before Pixls.us notified me of your last post.
I don’t have any negative comments to make about Vuescan/Ed/Beverly - it is after all a tiny organisation which does do a remarkably effective job. It seems to me though that the challenge of distinguishing between image defects and valid image pixels is precisely the kind of task LLMs are purposed to do. Given the tiny size of Ed’s operation, I do not expect him to implement an ‘AI’ technique in Vuescan any time soon.
Yes, it is quite noticeable, now that you bring it to my attention.
Is SilverFast going to be more effective?
Their approach is completely different. Although there are problems with Kodachrome despite their Kodachrome mode, what you see as corrections in the preview is also what is corrected in the final image, according to tests I made some time ago. I assume that each defect is corrected by interpolation, which is visible as a smooth appearance of where large defects were. You notice this also in the computing time: Many defects in an image take longer to correct.
It is important to not use the auto-setting iSRD and instead fine tune the parameters, especially the threshold. Here is an example with SF9 iSRD:
Defects detected are marked in red.
Upper left: threshold set to a minimum
Upper right: IR-channel optimized as described above
Lower left: automatic mode
Lower right: IR-channel
Hermann-Josef
Just for completeness, here is the corresponding difference image to the upper left case above:
Hermann-Josef
OK, I had some time and started investigating your kodachrome scans. This is pretty interesting, so thanks again for sharing! My findings so far follow, but I did not complete my investigation yet, so this is some early feedback …
First, when I import your image, it appears very dark in most parts:
The infrared layer shows a duplicate of the image, which is expected. This happens as well for non-kodachrome film, but typically the ghost image is more faint for non-kodachrome.
One can find a range where the dust/scratches become more prominent (see the spot in the sky) and only the darkest parts of the ghost image add to this mask:
A low threshold seems to work well for that purpose:
With a bit erosion and dilation, some more of the smaller sprinkles in the dark areas vanish:
However, shifting the image a bit to the left shows that not all issues are seen in the infrared image, e.g. the flare-like artifact is not recognized when compared to the image above:
Furthermore, there are a lot of extremely small defects that require a lower erosion and a higher threshold setting, which leaves much more of the dark parts of the image in the mask:
This is where the question from before becomes important. The contrast between dust and ghost image seems quite good in the brighter areas, only in the darkest shadows, the ghost image seems to dominate (see e.g. the sky, but also in the ghost forest there is some contrast between the ghost image and some darker spots:
As said, only some first findings, to be continued. If you or others have some answers, or maybe a second shot of another subject with different dynamic range borders, this would be great for further investigations.
The image appears dark since it is not gamma-corrected. This is the “raw” image from the scanner.
Hermann-Josef
OK, I probably wrote my question the wrong way. The question is, why is there no detail in the shadows? Is it because of the scan or did it already vanish while shooting, or is it even me importing the image into gimp?
@chris
I do not know what you expect, but here is my final image:
It pretty much is what I see in the slide with a magnifying glass on a light table.
Hermann-Josef
