Quality loss

[saliniero]

When it comes to exporting the photos, I see that I lose a lot of quality when I review them in JPG (I export them at maximum quality). People who upload photos to 500px for example, have a much higher quality (It is the same camera).  Where can the failure be?

[Jowday]

On 4/8/2020 at 11:23 PM, saliniero said:

When it comes to exporting the photos, I see that I lose a lot of quality when I review them in JPG (I export them at maximum quality). People who upload photos to 500px for example, have a much higher quality (It is the same camera).  Where can the failure be?

Serif re-used the poor JPG algorithms used in their legacy programs (The Plus range). I know. I checked. Output is inferior and not optimized for size either. Neither is it an option.

But I can assure you than when it is finally improved you will see release notes and marketing with headlines like "MASSIVELY IMPROVED JPG OUTPUT" 🤨

[R C-R]

On 4/8/2020 at 4:23 PM, saliniero said:

When it comes to exporting the photos, I see that I lose a lot of quality when I review them in JPG (I export them at maximum quality).

Keep in mind that even at maximum quality, standard JPEG compression is lossy (there is a lossless version but it has never been widely adopted).

The discrete cosine transform (DCT) algorithm the standard uses attempts to discard the highest frequency image content that is least noticeable to humans first & then progressively more of it as the quality is lowered. Since high frequency content corresponds to the amount of fine detail in the image,  this means that images with only small amounts of fine detail won't display much if any apparent quality loss if exported at maximum quality, while those with lots of fine details will.

Since fine detail in real photos includes not just the sharp color transitions characteristic of well focused photos but also luminosity & color noise originating in the camera sensor (which may be visually almost undetectable), it is not always obvious why a photo exported to JPEG at maximum quality does not look as good expected, particularly when compared to other photos uploaded to various web sites.

One reason for this is that some of those other photos may have used certain preprocessing techniques to remove some of their high frequency content before being exported to JPEGs. Two of these techniques are available in Affinity Photo. The Compression efficiency video tutorial  by @James Ritson explains how to use them.

[lacerto]

(...)

[R C-R]

11 minutes ago, Lagarto said:

I developed this in Photoshop and saved it as a TIFF file, then used the same file in both Affinity Photo and Photoshop to reduce it to 800 x 600px (Bicubic resampling in both cases) ...

So what are you actually testing, just the export to JPEG at 100% quality alone, or the combination of that and how each of the two apps applies bicubic resampling when downsizing images?

[walt.farrell]

14 minutes ago, Lagarto said:

As can be seen Affinity Photo has practically lost all the texture of the original image while still producing a larger file size. This, however, is not mainly a result of the JPG compression algorithm but softness left in the image by bicubic resampling, which produces much softer image than Photoshop's equivalent,

While that does seem to be true, if you want it sharper why not use one of Photo's better algorithms? Lanczos 3, either separable or non-separable, should give better results than bicubic.

[lacerto]

(...)

[lepr]

1 hour ago, Lagarto said:

There is a clear difference, indeed. I tested this by taking a photo of a painting, where the original texture of the canvas and painting looks like this:

 

I developed this in Photoshop and saved it as a TIFF file, then used the same file in both Affinity Photo and Photoshop to reduce it to 800 x 600px (Bicubic resampling in both cases), then exported to JPG using maximum quality in either apps. Photoshop produced a file the size of which was 417KB and Affinity Photo a file the size of which was 471KB, and here is a 300% enlargement of the results:

a) Photoshop:

b) Affinity Photo:

As can be seen Affinity Photo has practically lost all the texture of the original image while still producing a larger file size. This, however, is not mainly a result of the JPG compression algorithm but softness left in the image by bicubic resampling, which produces much softer image than Photoshop's equivalent, and I assume that the same algorithm is used in context of JPG export when the user specifies the export file size and downsampling is needed. In this case the original file size was 4,592 x 3,448px.

It is of course true that when these kinds of images are viewed at 100% zoom state, the differences are often insignificant. But in modern usage photos are very often zoomed in with mobile devices so details do count. As does of course the file size.

As mentioned, there was not any kind of preprocessing involved, so it does not seem likely that the better quality in the kind of 500px photos mentioned by OP could be explained by this. They probably just have been produced with better algorithms, both in relation to resampling, and compression. 

Are you sure you didn't use Bicubic Sharper in Photoshop? Last time I used Photoshop a few years ago, there was also "Bicubic Automatic" which does Bicubic Sharper when downscaling and Bicubic Smoother when upscaling.

You'll get a result similar to Bicubic Sharper by applying an unsharp mask filter after doing bicubic.

 

[Pompel]

Does this mean that the quality of images exported in Jpg (in AP) is poorer/lower than in other photo programs? (starting file is a TIF file), or have i misunderstood the topic/discussion?
What gives the best quality (Jpg): Lanczos 3 separable or Lanczos 3 non separable?

tss

[lacerto]

(...)

[lacerto]

(...)

[Pompel]

Hmm .... so I will never know if the image I am working on right now might have a better quality if I export it with a program other than AP? Unless I have another program to compare with.

Hmmm… again. Thought I had the best app😊.

Maybe I need to go shopping again, a secondary app.

 

tss

[lepr]

43 minutes ago, Lagarto said:

Yes, that's what I used so in practice Bicubic and Unsharp mask. One could expect Lanczos 3 Separable doing something like this, but no, Bicubic is very shoft, and the sharpening effect applied afterwards is not anywhere near the Bicubic Automatic of Photoshop. This image was not probably the best test for quality because it had a clear texture of its own, but I wanted to use it mainly because I wanted to see how compression works with this kind of image which itself has "disturbance".

Bicubic Sharper is not bicubic and unsharp mask in practice, so no wonder the results are not identical.

Without seeking a dispute, I want to point out that I posted because you wrote that Affinity's Bicubic produces much softer results than Photoshop's equivalent, and showed images to compare them. However, Photoshop's equivalent to Affinity's Bicubic is Bicubic, not Bicubic Sharper, so your images weren't actually comparing equivalent resampling methods. 

[lacerto]

(...)

[lepr]

15 minutes ago, Lagarto said:

[...] Bicubic Auto (or Bicubic Sharpen [sic] which I have no experience because I'm using Photoshop CS5) [...]

You do have experience of Bicubic Sharper because that's what Photoshop uses when downscaling with Bicubic Automatic and you said you used Bicubic Automatic.

15 minutes ago, Lagarto said:

As could be seen, assuming that "Bicubic" generally behaves similarly as Photoshop's "Bicubic" resampling algorithm is a false assumption but one that many are likely to make.

You have not been comparing Affinity's Bicubic with Photoshop's Bicubic. You have been comparing Affinity's Bicubic with Photoshop's Bicubic Sharper. Photoshop's Bicubic and Photoshop's Bicubic Sharper are not the same.

[lacerto]

(...)

[ashf]

Photoshop applies unsharp filter automatically without asking user when resize the image.
(This means multiple time of resize may spoil the image on Photoshop.)
So in order to get similar result, you need to apply unsharp filter after resize the image.

So just comparing resized image is not proper testing.

[ashf]

This is what Affinity needs

 

[heavyside]

On 4/11/2020 at 2:42 PM, Jowday said:

Serif re-used the poor JPG algorithms used in their legacy programs (The Plus range). I know. I checked. Output is inferior and not optimized for size either. Neither is it an option.

But I can assure you than when it is finally improved you will see release notes and marketing with headlines like "MASSIVELY IMPROVED JPG OUTPUT" 🤨

Interesting. I can only assume this claim is true (or at least very close to the mark) based on my own findings when trying to export to a JPG. Can we please get some feedback from Serif about the state of JPG exporting in Designer/Photo?  Is it possible we will get improved algorithms any time?

There have been numerous comments/complaints about the lack of control (no real-time export preview window - one of photoshops best features!), the sub-par quality of of the resulting images regardless of the settings you use (yes even at 100% they are terrible), and also the horrendous file-sizes for those of us intending on using the images online (with quality set to 100%  - a 600x250 image was exported at 200kb and it looked atrocious. For the image I got, I would expect it be closer to 50kb at absolute max). 

I have found it literally impossible (as in, I am yet to achieve it) to get an export result that I am completely satisfied with.  

Perhaps I am doing something wrong, but it seems to be an issue that others are also dealing with. 

Serif, please help! I love your software, but hate the prospect of exporting from it! 

 

Kind Regards, Sam.

[Jowday]

4 hours ago, heavyside said:

Interesting. I can only assume this claim is true (or at least very close to the mark) based on my own findings when trying to export to a JPG. Can we please get some feedback from Serif about the state of JPG exporting in Designer/Photo?  Is it possible we will get improved algorithms any time?

Serif PhotoPlus X7 vs Affinity Photo

Other comparions:

 

[ashf]

I  think Serif should just use mozjpeg, established jpg encoder that's superior than Photoshop.
https://github.com/mozilla/mozjpeg

[ljredux]

Problem is, there are so many file formats and most of them seem to be problematic.  I'm trialling AP and spent all my free time over two days trying to get better results with resizing (resampling) and exporting to various formats, but Photoshop simply does it better (whatever the format) with the exact same files.  I have more control over the exporting process too.  I hate Gimp but even that does better with resampling and exporting.  Maybe I'm expecting too much from software at this price point..., I don't know.

[walt.farrell]

1 hour ago, ljredux said:

but Photoshop simply does it better (whatever the format)

Some examples of what you're finding might be helpful, including screenshots. But one thing I've read that might be relevant is that Photoshop by default applies some sharpening when you export. Affinity Photo doesn't; it tries to export the image you have composed, without making assumptions about what you might want.