Resize quality

[Danel]

Hello ,
I recently did a comparison of resizing a 6000x4000 photo to 2048x1365: 50% quality with 6 different programs.
It turns out that Affinity is far behind the other five because of a very bad management of the moiréing and the definition.
It's a shame because I have to use another software for this specific operation.

I use "Document/Resize".
Is there any way to improve this?

(PS we have to do this work on many photos)

 

[R C-R]

Does "50% quality" mean you are exporting the photo to the JPEG format or are you doing something else?

[NotMyFault]

You can use several different resample methods.

It totally depends on the type of source image and your artistic intention which method delivers best results.

• pixel art?
• Line art?
• Photos?
• Very fine details (already existing morie)

So if you can give an example file and intended purpose / acceptance criteria we might find the best method.

Your resize non-integer factor (6000:2024) might cause additional quality loss, too

[Danel]

Thank you for your answers,
We are a photo club and following the pandemic, we use Skype instead of our local meetings.
 Our facilitator receives our images, and show them to us online.
The criticism and exchanges can begin.
This works perfectly for over a year.
For the display and file size adjustments, we found it reasonable to decrease the size of our photos, and to increase the file compression when sending our images.
After several tries, it turned out that the 2048 pixel format with 50% quality allowed us to keep a lot of details of the photo and to stay in a very acceptable weight range, between 200 and 400Kb depending on the type of image.
It is here that we discovered a difference in results, quite small between most of the software, but strong with Affinity, which has a combination of sharpness and moiréing problems.

[telemax]

If you're still talking about JPG compression at 50, then I recommend using a value of at least 90 (for Affinity apps).

[Danel]

 

Thanks Telemax,

I'm afraid I didn't explain myself well: we decided to compress at 50% because it's an acceptable compromise to have smaller files while keeping a correct quality.
It should not be forgotten that as there are many of us, and as each one proposes between 5 and 10 photos depending on the case, the volume of the transfers becomes very important.
I repeat that 50% and 2048 is still very acceptable for most other softwares...

[thomaso]

18 hours ago, Danel said:

a very bad management of the moiréing and the definition

Is there a specific type of image content? Have you tried the different resample methods in Affinity, in particular the two Lanczos options?

Additionally, to reduce the file size of a jpg exported in Affinity you could try the ImageOptim.app  https://imageoptim.com/mac
It has its own quality setting, so maybe you choose a high value for the export from Affinity and target the 50% with ImageOptim only, else 2x 50% might reduce the quality more than expected.

[John Rostron]

I recall reading that the "quality" percentage means different things in different programs. It would be better to specify a maximum size for the exported file. This would seem to be more relevant to the purpose that you describe.

John

[R C-R]

1 hour ago, John Rostron said:

I recall reading that the "quality" percentage means different things in different programs.

FWIW, I read somewhere long ago that Photoshop in particular once (& maybe still does?) uses a JPEG 'quality' scale that does not directly relate to the percentage values typically used in other software, at least when using the "Save" or "Save as" export settings. (The now depreciated "Save for Web" function does, or did -- I am not sure about that either.) The "Save" or "Save as" scale goes from 1 to 12, but it isn't linear so it is difficult to compare it to apps that use a linear percentage scale.

Anyway, like @thomaso suggested, if you are a Mac user the free ImageOptim app works very well to reduce JPEG (& PNG) file sizes.

[Danel]

Of course, software developers do not indicate the algorithms they use....
Finally, what interests us at the club is to have images that are correct enough to be displayed but that are not much larger than 400Kb, regardless of the process used,  (which is what most other software allows us to do).

So we use Faststone but it is a shame that those of us (as me) who use Affinity cannot directly process their files like other software.

I'm afraid I don't have a simple solution for this little problem.
Thanks to all

[thomaso]

25 minutes ago, Danel said:

I'm afraid

@Danel, you are not very verbose in terms of precise or detailed with information

For instance we (just me?) neither know yet what this "bad management of the moiréing and the definition" means to you respectively to the images, nor this "simple solution" or that "little problem". Also, you don't seem to respond to specific suggestions, which makes it hard or even impossible for the community to satisfy your expectations.

Now, since your group prefers using "Faststone" (apparently free?) – why don't you use it, too?

[thomaso]

Still just a guess: If you are looking for a software to batch convert images, you might be interested in "xnConverter". https://www.xnview.com/de/xnconvert/

Honestly, I cannot judge whether a different workflow than the one you are currently using feels "simple" to you, I can only say that a batch processing process always feels attractive to me as soon as more than one file is to be processed simultaneously with the same settings.

[David in Яuislip]

1 hour ago, Danel said:

Finally, what interests us at the club is to have images that are correct enough to be displayed but that are not much larger than 400Kb,

that's fair enough so all club members should use the same procedure regardless of what software they have used to process the image from the camera, ie I use PhotoLab so I could use that to produce a 2048 wide image, say 100% jpeg, then run that through FastStone with the agreed compression. I think comparing the output of various programs using the same quality value will get nowhere

 

19 hours ago, Danel said:

It is here that we discovered a difference in results, quite small between most of the software, but strong with Affinity, which has a combination of sharpness and moiréing problems.

It would be very interesting to see an example of an original and an Affinity processed result showing such problems

On a sort of similar note I started to look at the effect of compression for web images with the intention of minimising file sizes. The problem I find is that there is no way of quantifying the perceived quality. You can look at the graph and see a point of diminishing returns but if that doesn't produce an acceptable image then so what, the quality needs increasing? Anyway, it's not that bleak, I have settled on 40% webp's, software is ImageWorsener

[Danel]

Thank you David for this information.
Thomaso will also find an example below.

 Attached is a comparison of the same "JPG 100%" photo, reduced to a comparable size:
On the left Affinity: 70% size 511KB, on the right Faststone 50% size 501KB.

I used faststone, but I could have shown Rawtherapee (which is a bit better), or Gimp or Daktable, Lightroom.

Finally, the purpose of my post was not really to compare different software, but only to know if there was a setting on Affinity that I didn't know about and that would allow my friends to get a good display quality during our Skype sessions: that's all.

I thank you all, for the attention you have given to my message.

[NotMyFault]

There is something wrong. The right left jpg shows severe artifacts, e.g. right of hat, above chair.

It seems the file was misaligned by 1/2 pixel on y-axis, everything is blurred in y-direction.

maybe bilinear is not the best resample method in this case.

if you can provide one uncompressed source image (png format), we can check which export settings give good results.

Edited November 7, 2021 by NotMyFault
Corrected left/right

[GarryP]

NotMyFault: Can you explain what you mean by “severe artefacts” on the right-hand image?
The right-hand image looks quite a lot sharper and better, overall, to me and I can’t see any problems with it, certainly not anything “severe”.
Note: I am viewing the image after clicking on it, not just while browsing the thread.

[thomaso]

On 11/5/2021 at 4:11 PM, Danel said:

bad management of the moiréing and the definition.

In your sample the Faststone version is sharper than Affinity but shows more moirè (which appears consequent because of more sharpness). Also consider that moiré is related to hardware pixel size of the displaying monitor and therefore varies with zoom factor.

Sharpness (="definition" ?): Your samples don't tell which resample method was used for each. A quick test with resampling the Faststone result shows obvious differences on Affinity export between bilinear, lanczos3 and lanczos 3 non-separable: (click to enlarge, <> to navigate)

Nevertheless, and no doubt, Affinity exports appears to result in larger file size than other apps might do, the resampling method influences the file size, too.

Therefore the method to export rather uncompressed and do the compression in another app became mentioned in posts above. If Affinity is used only for cropping then possibly it would not get used at all in such a workflow. So, depending on your entire workflow with the images (incl. how many totally, how many in 1 session) batch processing may or may not feel as "simple" to you, regardless of its low effort for setup + export all with 1-click.

According to its metadata your comparing sample JPG seems to be created by "PhotoFiltre 7" – which may alter your results of Affinity & Faststone in any or in different ways.

On 11/5/2021 at 4:11 PM, Danel said:

I use "Document/Resize".
Is there any way to improve this?

What I haven't tested yet is whether resizing within the opened document or resizing during export does make any difference. Theoretically I wouldn't expect a difference but Affinity did and does surprise in its behavior and results occasionally in a few of its features. So who knows, maybe for certain image content or within certain resizing values only?

EDIT: "PhotoFiltre 7" seems to be for Windows only, like "FastStone" - this makes me wonder again: Why don't you prefer using FastStone, like the others in your group, at least for better comparability? Is it more a general interest in "how to get more from Affinity" / "why does Affinity work the way it does"?

Edited November 7, 2021 by thomaso

[thomaso]

1 hour ago, NotMyFault said:

provide one uncompressed source image (png format)

I wonder whether the resample setting for all export pixel formats in Affinity has an influence if the image does not get resized on export. If it does nothing I would expect its menu to be grayed-out as long the size values for export are not altered by the user.

[R C-R]

2 hours ago, NotMyFault said:

There is something wrong. The right left jpg shows severe artifacts, e.g. right of hat, above chair.

Hmmm. I see only one comparison JPEG in the above post & for me both sides of it look essentially the same, both when viewed in a separate Safari window & when downloaded as a 2560 × 1374 px JPEG & opened in the Mac Preview app.

What am I missing?

[thomaso]

11 minutes ago, R C-R said:

What am I missing?

I don't see this specific artifacts, too, but, viewed in 100 %, the Affinity result (left) appears obviously softer / less sharp then the Faststone (right). Whereas exporting a detail of the right half in Affinity makes it appear quite the same with lanczos resampler and looking a bit overdone with lanczos-non-separable (each with 70% quality).

[R C-R]

8 minutes ago, thomaso said:

I don't see this specific artifacts, too, but, viewed in 100 %, the Affinity result (left) appears obviously softer / less sharp then the Faststone (right). Whereas exporting a detail of the right half in Affinity makes it appear quite the same with lanczos resampler and looking a bit overdone with lanczos-non-separable (each with 70% quality).

Weird. Like I said, for me the two sides look essentially the same when downloaded in Safari & viewed in the Mac preview app. Neither side looks soft; both sides look very sharp.

This is the 1.8 MB file I get: 

[NotMyFault]

Ok, to better explain how i compare:

• used that jpeg image with both files side-by-side
• duplicated layer and aligned them so both are atop of each other (at the right half of document)
• zoom in to pixel peep (16x)
• Toggle the middle layer on/off. The difference in sharpness is striking. You won’t see it when zoomed out.
• Tried blend mode difference (top layer), it shows that almost all fine details are missing in Affinity export version.

compare.afphoto

[User_783649]

27 minutes ago, R C-R said:

Weird. Like I said, for me the two sides look essentially the same when downloaded in Safari & viewed in the Mac preview app. Neither side looks soft; both sides look very sharp.

Very obvious and noticeable difference in sharpness for me between these pictures. Overall quality and sharpness is better on the right side. Especially parts with hat fibers and chair mesh material. However, it looks like some amount of output sharpening was applied to image on the right. Looks like a bit too much sharpening. Image on the left definitely looks more compressed to me, and lacks of details.

Again, it would be useful if @Danel could tell us how the export from Faststone was made exactly? Are any extra-settings or options were involved?

[User_783649]

It's nice we have resampling options to choose from in Affinity Photo.
However, we should understand the difference between them so we can use them on purpose.

I made another test. You can repeat it by yourself.

test_original.jpg (4.9mb)
Original image. Open it and export with 1000px height, jpg, 50% quality, bilinear, bicubic or lanczos.

Spoiler

test_1000h_bilinear_50q.jpg (172kb)
Obviously, it blurs image a little bit and some details are being lost. Probably good to use in app, for canvas performance reasons, but totally not suitable for exports.
Most likely, that @Danel used exactly this (default!) option, so that might probably explain why the result was not on par with other software.

Spoiler

test_1000h_bicubic_50q.jpg (204kb)
Seems like an optimal option in most cases. Not blurry and not over sharpened.

Spoiler

test_1000h_l3s_50q.jpg (217kb)
A bit more sharpness and details than in Bicubic. Might be a good choice when some extra sharpness is needed.

Spoiler

test_1000h_l3n_50q.jpg (260kb)
Very aggressive sharpening. Expected for this type of Lanczos.

Spoiler

Final thoughts:

It would be better for Serif to change the default resampling method in Affinity Photo export interface to Bicubic.

[Danel]

1 hour ago, Alex M said:

 @Danel could tell us how the export from Faststone was made exactly? Are any extra-settings or options were involved?

No, nothing:
The original 6000x4000 photo was loaded into Faststone, then directly reduced with the reduction tool to 2048 and then saved to my desktop at 50% quality.

I think everyone could try it for themselves as Faststone is free, not heavy software and can be uninstalled quickly.