Better resampling filter to combat blurry previews

Propheticus

• September 14, 2022 15:51

This has been asked before and issues have been reported before. A community search for the words "blurry preview" should find you plenty of examples.

I tried to make an apples-to-apples comparison and came to the conclusion a sharper preview should be possible even without applying sharpening at all. 
It is understood certain image processing, like sharpening and noise corrections, might not be fully performed until you zoom in. This can be explained as a quality/performance trade-off, where applying all processing while zoomed to fit would slow down the experience too much. 
Dragging the sharpening slider does seem to have an effect while zoomed to fit, but the image is far blurrier than expected.

When I check the size of Capture Ones preview files for a setting of 2560px (to match display width), I find a ~1 to 1.5MB file size.
I looked at an example photo and found a 1.04MB preview .cop file. I then created a JPEG export with 2560px width and picked a JPEG quality to get the same file size. Quality 87 results in 1.03MB. Sharpening amount for the export was set to 0(!) and then reset to defaults (120).

Then, I compared a 51% scaled version of this export (JPEG, Q87, no sharpening) in XnViewMP (Lanczos scaling) versus the preview in a windowed Capture One to match the image width on screen. Even with sharpening in C1 at default, I found the unsharpened export to be sharper.

My conclusion: something can be greatly improved in scaling/resampling of the image. 
In a full screen viewer without toolbar the issue seems less pronounced, but still the unsharpened export to me looks slightly sharper than the sharpened C1 preview.
Even in full screen, scaling is needed because the screen ratio of 16:9 is wider than that of most photos. To fit the height, a 2560px wide image is still scaled down to ~75-80%.

The example, unsharpened jpeg above, sharpened C1 preview below:

A lot of resources (and debates) can be found online about what are the best image resampling filters. Windowed sinc/jinc filters are best for downscaling real-world images. Lanczos is often mentioned as a good option when balancing quality and performance. Another is Catmull-Rom, which essentially is a cheaper to calculate cubic filter with results that almost match a 2-lobed Lanczos. [best source I found]

 

[related topic]

5

• 

  Noob with a Nikon

  • September 14, 2022 16:09

  If I understand the findings of your experiments correctly, an image in the C1 main viewer is less sharp than the same image exported to a JPG at 87% quality?

  0
• 

  Propheticus

  • September 14, 2022 17:34

  Indeed. Export was 2560px wide, sharpening 0, quality 87.
  This was sharper than C1 viewer (preview) of the RAW with sharpening 120.

  0
• 

  Class A

  • September 14, 2022 19:37

  I have requested better quality previews years ago and even had exchanges with the developers, so you probably better not get your hopes up.

  For some reason Capture One (the company) seem to think the previews are good enough at the current level.

  One could theoretically work using "Proof" Mode at the desired output resolution, but I find the label across the image too distracting and zooming at 100% obviously doesn't work anymore.

  1
• 

  Propheticus

  • September 15, 2022 18:17
  • Edited

  Did not see any improvement when choosing a proof profile (icc), neither when enabling recipe proofing of the Full res HQ recipe. Then made a new export recipe 2560px wide, quality 98 and AdobeRGB icc.
  Now enabling recipe proofing does the trick. So; a 2560px wide preview is blurry, but proofing a 2560px wide recipe works fine... right.

  Thanks for the tip! I'll keep this dummy recipe just to be able to assess sharpness while viewing the entire photo.

  1
• 

  Propheticus

  • November 17, 2022 18:37
  • Edited

  Found a related issue caused by strange scaling methods. In images with fine details like foliage the zoom level influences perceived saturation. The way details are averaged/combined on downscaling somehow messes up the colour.
  Export recipe proofing does not fix it and proofed colours do not match the actual output.

  At 33%, over saturated (33% zoom of exported JPEG in XnViewMP below):

  

  

  At 50%, under saturated:
  

  At 67%, correctly saturated:
  

  (Recipe proofing was on in the 3 examples above and XnView is set to use the monitor ICC profile)

  The zoomed to fit up till 33% zoom is clearly too saturated. When comparing the recipe proofed preview side by side with the exported JPEG in an image viewer that does proper downscaling the difference is very obvious.

  1

Post is closed for comments.