Light Gathering, noise and diffraction: an XH2 question

[MrAndreas]

If the RAW processing pipeline includes a color filter array (CFA), there has to be interpolation (guesswork) in reconstructing the image. The more image data there is to work with, the fewer errors there will be. Those errors are typically so small that they aren’t immediately apparent, but upon closer inspection various errors and artifacts become obvious - both in color and detail.

Below is a comparison of the 16MP X-T1 sensor vs. the (downscaled to 16MP) X-T5 sensor and then both viewed at 200% to exaggerate the situation - all the obvious false detail/artifacts that can easily be seen in the X-T1 image text is due to demosaicing errors - processing guesswork that failed. With 40MP of image detail to work with, the X-T5 image contains fewer errors, and while there is a loss of fine detail (resolution) when downscaled, the more faithful detail rendering is still retained, producing a better looking image at 16MP despite its resolution being significantly reduced. Some will say that, hey, you’re pixel peeping to see those errors, and they would be right, but it’s those low level errors that get exaggerated with sloppy sharpening …“worms” and ugly foliage especially, that can easily be noticed at normal viewing sizes, and that are largely absent with the 40MP sensor - even with Adobe’s less than stellar standard X-Trans demosaicing.

Thank you for the detailed explanation. Bookmarked it.

While I never look at images at 200%, the "ugly foliage" phenomenon is observable on the OOC JPEGs. Demosaicing RAF in darktable with a proper algorithm, properly configured, helps a lot.

 

[Erik Baumgartner]

If the RAW processing pipeline includes a color filter array (CFA), there has to be interpolation (guesswork) in reconstructing the image. The more image data there is to work with, the fewer errors there will be. Those errors are typically so small that they aren’t immediately apparent, but upon closer inspection various errors and artifacts become obvious - both in color and detail.

Below is a comparison of the 16MP X-T1 sensor vs. the (downscaled to 16MP) X-T5 sensor and then both viewed at 200% to exaggerate the situation - all the obvious false detail/artifacts that can easily be seen in the X-T1 image text is due to demosaicing errors - processing guesswork that failed. With 40MP of image detail to work with, the X-T5 image contains fewer errors, and while there is a loss of fine detail (resolution) when downscaled, the more faithful detail rendering is still retained, producing a better looking image at 16MP despite its resolution being significantly reduced. Some will say that, hey, you’re pixel peeping to see those errors, and they would be right, but it’s those low level errors that get exaggerated with sloppy sharpening …“worms” and ugly foliage especially, that can easily be noticed at normal viewing sizes, and that are largely absent with the 40MP sensor - even with Adobe’s less than stellar standard X-Trans demosaicing.

Thank you for the detailed explanation. Bookmarked it.

While I never look at images at 200%, the "ugly foliage" phenomenon is observable on the OOC JPEGs. Demosaicing RAF in darktable with a proper algorithm, properly configured, helps a lot.

Yeah, while I’m a Lightroom user, I’ve been using Iridient X-Transformer for demosaicing X-Trans files since day one with my Fuji cameras. Adobe’s enhanced A.I. demosaicing/NR is fine with X-Trans now, but the standard demosaicing is still subpar (though much less problematic with the higher resolution sensors).

 

[Lance St. Paul]

Another way to look at diffraction is that instead of one sharp plane of focus it becomes more diffused spreading a sharpish zone further, simply not as peaked at one distance. It can be useful. It is not inherently evil.

Beetlejuice

 

[Morris0]

Using the XH2 with a long lens at apertures of F8 and smaller may lead to diffraction and noisy images that even when cropped my not give you the best results, largely due to a poor signal to noise ratio. If this is generally accurate, and I am not saying it is exactly so, when purchasing a lens whether prime or zoom would it not be to one's advantage (again generally speaking) to use the fastest lens that meets your use case?

Not necessarily. People often treat diffraction as some magic aperture value where an image is unusable. When it's really a slow, generally imperceptible decline in sharpness.

I've read comments from folks saying they never go past f/4-5.6, etc, for maximum sharpness. That's absurd. Because one, it's not that big a deal. And two, depth of field gains will often increase perceived resolution, despite diffraction losses. Meaning, if I shoot a telephoto landscape at f/4 to avoid diffraction, the image foreground may still look soft because sections are out of focus. If I use f/8-11 instead, it will look better, despite hitting diffraction territory.

Diffraction isn't that big a deal; use whatever aperture value best fits your creative intention, knowing it comes at the cost of max resolution. And compensate in post with added sharpness, if needed. Capture One has a Diffraction Correction tool that's very handy to apply at the start.

I agree, while the image may have less micro contrast affecting the appearance of sharpness, good photos can be taken at any aperture.

Morris