Layer/Image scaling quality AP vs PS

[MancDan]

Hi, just noticed this.

Trying to run through an old job done in PS using AP instead. I'm trying to gauge how quickly I can work in AP in comparison to PS.

I've attached two images. Both attachments have a sun flare image set to screen mode (or add in 32 bit, tried both). This sun glare texture has been scaled up, both in PS and AP the same %. There are no smart objects used in the PS version.

As you can see, the quality of the image is a lot worse in AP than it is in PS, very pixelated, yet they have both been scaled identically the same amount. Am I missing something? We cannot switch to AP if this is the result, we'd be stepping backwards.

Any ideas? Image quality is set to best in performance.

Edit: it appears the images are scaled differently in the example below, this is just because of zoom extents in PS and AP and how it positioned when i took the screen shot. Also I would of course expect some pixelation blowing up an image like this, but the difference between programs is pretty stark.

Thanks

[Gabe]

Hi @MancDan ,

Whtch resample method have you used? 

Resampling methods

The following resample settings are available:

• Nearest Neighbor—simple resampling which has the fastest processing time. Use for hard-edge images.
• Bilinear—algorithmic resampling for use when shrinking images.
• Bicubic—algorithmic resampling for use when enlarging images. Resampling is smoother than Bilinear but has a slower processing time.
• Lanczos 3—complex algorithmic resampling which gives the best results but with the longest processing time. Available as 'separable' and 'non-separable'; the latter gives marginally better results, but is slightly slower than 'separable'.

Thanks,

Gabe. 

[MancDan]

Hi

 

I only have two options, Bilinear and Nearest Neighbor. It was set to Nearest Neighbor actually, but I switched to Bilinear and restarted. Still looks really pixelated compared to PS.

[Gabe]

I suggest you either resize it using Document > Resize document, or export using one of the "smooth" algorithms, like Lanczos or Bicubic. If you just preview it in the working area, that will be treated as either Bilinear or Nearest Neighbor and does not reflect the actual exported file, unless you use the same algorithm to export as you use for preview. 

[MancDan]

27 minutes ago, GabrielM said:

I suggest you either resize it using Document > Resize document, or export using one of the "smooth" algorithms, like Lanczos or Bicubic. If you just preview it in the working area, that will be treated as either Bilinear or Nearest Neighbor and does not reflect the actual exported file, unless you use the same algorithm to export as you use for preview. 

There is no difference exporting with either of those algorithms, it still saves out pixelated.

[Gabe]

Can you attach the afphoto project file so we can have a look? Also, please attach the original images, unaltered

[MancDan]

Attached.

You notice it as you zoom in. It has been scaled quite a lot, but the difference between the PS scale quality and AP scale quality is clear.

Test_AP_00.zip

[MancDan]

Just now, MancDan said:

Attached.

You notice it as you zoom in. It has been scaled quite a lot, but the difference between the PS scale quality and AP scale quality is clear.

Test_AP_00.zip

In fact you can see it zoomed out quite clearly. PS, is pretty smooth

[MancDan]

A slight Gaussian blur helps, but not ideal.

[Gabe]

Right. You're scaling that image up 6 times, and pixelation is expected. Not much we can do, as it's not a bug. Different software have different algorithms to resample. Below is an example from 

PS: Automatic

And AP: Lanczos 3 Separable

 

[MancDan]

Yes I made it pretty clear it was scaled up quite a lot, its a pretty extreme example and pixelation is expected. The point is, I don't have the pixelation issue with PS and I wondered why. Thanks for your answer though.

So,  were putting it down to how PS and AP re-sample imagery. Got it.

My worry is, using less extreme scaling examples, will such a gap in quality be noticeable too.

[MancDan]

Also, the issue much more noticeable when using layer modes

[Medical Officer Bones]

Did some testing, and in other applications scaling up with Bicubic Spline results in an even nicer/usable scaled up version than your Photoshop version.

Unfortunately, Affinity Photo doesn't give the user a choice in resampling algorithms when upscaling and rasterizing a specific layer.

I then attempted to upscale the original with the Resize Document option, and selected Bicubic as the resampling method. The result leaves a LOT to be desired of compared to any of the other apps I tested in (Krita, PhotoLine, Gimp).

In my opinion, based on my previous experiments with downscaling assets, and now with this particular upscale example, Affinity's resampling code base needs to be looked at again. It is a quite fundamental thing to get right. If other applications have no issues with up- and down-scaling, my only conclusion can be that something is amiss with Affinity's basic approach in resampling. The choices of resample methods in the preferences and document size dialog are fairly limited as well. For downsampling, for example, Lanczos and Bicubic are far from ideal choices, but no alternatives such as CatmulRom or MitchelNetravali are on offer in Photo.

As it stands, Photo is lagging behind all the competition in this area. Let's hope the devs will address this in an upcoming release, because it needs to be.

[Fixx]

About all serious work I do in AP (compositing a bunch of images) includes resampling so this is important issue. Though I usually let export handle resampling which seem to keep quality reasonably good. These are hi resolution images though.

[MancDan]

17 hours ago, GabrielM said:

Right. You're scaling that image up 6 times, and pixelation is expected. Not much we can do, as it's not a bug. Different software have different algorithms to resample. Below is an example from 

PS: Automatic

And AP: Lanczos 3 Separable

 

Also, in your example, have you scaled the image up 6 times, or have you just opened the image up in both PS and AP? If so, they aren't dissimilar initially, you need to scale up both images by the same percentage to notice the issue.

[MancDan]

I've just realized you set the re-sampling method to separable, not non separable this provides a much better result.

 

Thanks!

[Gabe]

43 minutes ago, MancDan said:

Also, in your example, have you scaled the image up 6 times, or have you just opened the image up in both PS and AP? If so, they aren't dissimilar initially, you need to scale up both images by the same percentage to notice the issue.

I have resized them to proportionally to 3100px on the long edge.