Paperwhite 2018 looks worse than 2013? (Picture inside)

[JSWolf]

Originally Posted by ilovejedd

The "colors" on the Libra is actually very close to what I see with the naked eye. It really is noticeably lighter compared to the Paperwhites. I find the Kindles do a better job with the grayscale conversion with regards to compensating for the limitations of the e-ink display.

Mind, the native KFX does look a bit better than the AZW3 version.

Will check out KOReader on the Libra later. Lol, was just too lazy to figure out how to assign button mapping on the Libra so both buttons go next on KOReader. Alas, only 8GB storage means frequent content management when dealing with manga.

Of the three images, I find the Libra to have better visible detail. The Kindles (in some places) have better black. But overall, the finer details such as the belt (above the rope) and her ear have noticeably better detail. The Kindles are a bit muddy in comparison.

I'd like you to try the following (if you don't mind). Take this image and make it a cover for an ePub. Send that ePub to the Libra with the option to dither it as grayscale. Copy the image off of the Libra and view it on each device and see how it looks. There should be no dithering from any of the devices as it's already dithered for eInk. That would show how the different screens look.

[ilovejedd]

Originally Posted by Gametaku

But earlier you mentioned the Libra has better contrast than both PWs?

Normal text. I expected it would be the same for graphics. Sadly, I was mistaken.

By the way, the PW3 display is better than I remembered. Now that I've had the chance to look at it again, I'd rank them as follows:

PW3 > Libra H2O >> PW4

[BookCat]

Please explain what "dithering" means in this context.

[JSWolf]

Originally Posted by BookCat

Please explain what "dithering" means in this context.

Dithering is a color reproduction technique in which dots or pixels are arranged in such a way that allows us to perceive more colors than are actually used. This method of "creating" a large color palette with a limited set of colors is often used in computer images, television and the printing industry.

It lets us perceive more shades then exist given that the eInk display only has 16-shades.

[NiLuJe]

@BookCat: An eInk screen can only visibly display 16 shades of gray.

If you don't do anything particular, that means, for every shade that isn't exactly those, it'll take a (mathematical) knife, and choose instead one of those 16 colors (the closest to the actual, full range value).
i.e., quantization.
In practice, it'll mean stuff like gradients will change color very abruptly. The type of artifacts that creates is called "banding".

Dithering involves maths shenanigans to reduce the *actual* amount of different colors in the image to those same 16 colors, hopefully with the least amount of fidelity loss when perceived by the human eye. The screen can then just display them as-is.

In practice, on a Forma: https://github.com/koreader/koreader...ment-459955126

Original, full color: https://github.com/koreader/koreader...ment-459877277
Original, grayscale: https://github.com/NiLuJe/FBInk/pull...ment-580505556
Properly dithered: https://github.com/koreader/koreader...ment-459958068

I recommend viewing those with no zoom, because scaling algorithms will smooth dithering over.

(Some context: The Clara/Forma/Libra (& theoretically the PW4/KT4/Oasis 2/Oasis 3) can provide a fairly decent dithering algorithm for free, because what serves as as "GPU" in these things can now do it properly. Kobo is taking advantage of that fact. AFAICT, Amazon doesn't, possibly because they've been handling it differently for quite a while (i.e., preprocessed content, or on-device software dithering)).

[ilovejedd]

Originally Posted by JSWolf

Of the three images, I find the Libra to have better visible detail. The Kindles (in some places) have better black. But overall, the finer details such as the belt (above the rope) and her ear have noticeably better detail. The Kindles are a bit muddy in comparison.

Alas, the loss of detail/muddiness on the Kindles is a casualty of either the scanning process or JPEG compression. Both the ear and belt actually look fine to my naked, nearsighted eye.

The Libra lost light shaded detail on the moon (above Beat). It does a better job of dithering on the moon glow in this case, though. I notice some banding on the PW3 and 4.

Originally Posted by JSWolf

I'd like you to try the following (if you don't mind). Take this image and make it a cover for an ePub. Send that ePub to the Libra with the option to dither it as grayscale. Copy the image off of the Libra and view it on each device and see how it looks. There should be no dithering from any of the devices as it's already dithered for eInk. That would show how the different screens look.

I typically don't use send to device on Calibre. Is dithering an option on the default Kobo device driver?

That said, the Libra H2O and Paperwhites have different resolutions so the image will likely be resized/resampled and then re-dithered by the Kindle. Besides, can't figure out a way to make Manga EPUBs and AZW3s for use with native reader short of using KCC which applies significant processing and cropping.

In any case, best bet is likely to use CBZ with KOReader on both.

[Gametaku]

Originally Posted by ilovejedd

Normal text. I expected it would be the same for graphics. Sadly, I was mistaken.

Could you do a text comparison between Libra and the PWs (if its not too much trouble)?

[NiLuJe]

@ilovejedd: It's an option for the *cover* being generated for the sleep screen (in the Kobo driver). (Also applies to the rest of the Library thumbnails).

(Which is why what Jon proposed is so roundabout. If you have ImageMagick on hand, you can do the same thing manually [-grayscale Rec709Luma -ordered-dither o8x8,16] ).

Other than that, no-one has ever bothered with adding any sort of image processing during the conversion/upload pipeline, probably with good reason. (i.e., there are much better tools for the job (KCC?), because you usually want something very very specific depending on the device, and that wouldn't gel well with the rest of what Calibre's doing anyway).

[j.p.s]

Originally Posted by NiLuJe

@BookCat: An eInk screen can only visibly display 16 shades of gray.

If you don't do anything particular, that means, for every shade that isn't exactly those, it'll take a (mathematical) knife, and choose instead one of those 16 colors (the closest to the actual, full range value).
i.e., quantization.
In practice, it'll mean stuff like gradients will change color very abruptly. The type of artifacts that creates is called "banding".

Dithering involves maths shenanigans to reduce the *effective* amount of different colors in the image to those same 16 colors, hopefully with the least amount of fidelity loss when perceived by the human eye. The screen can then just display them as-is.

In practice, on a Forma: https://github.com/koreader/koreader...ment-459955126

Original, full color: https://github.com/koreader/koreader...ment-459877277
Original, grayscale: https://github.com/NiLuJe/FBInk/pull...ment-580505556
Properly dithered: https://github.com/koreader/koreader...ment-459958068

I recommend viewing those with no zoom, because scaling algorithms will smooth dithering over.

(Some context: The Clara/Forma/Libra (& theoretically the PW4/KT4/Oasis 2/Oasis 3) can provide a fairly decent dithering algorithm for free, because what serves as as "GPU" in these things can now do it properly. Kobo is taking advantage of that fact. AFAICT, Amazon doesn't, possibly because they've been handling it differently for quite a while (i.e., preprocessed content, or on-device software dithering)).

I think you meant increase the effective number of different colors.

Attached is a 4 bit undithered conversion of your 8 bit grayscale image forced to 8 bit in case anybody has a broken PNG renderer. This shows what the image would look like on a 16 shade display with no dithering.

pngtopnm dither_test_gs.png | pnmdepth 15 | pnmdepth 255 | pnmtopng -force > undithered4bit.png

[BookCat]

Thank you for explaining dithering. In the case of manga I don't think sharp transitions between shade grades matter; manga tends to be an unsubtle art. I imagine what really matters to the OP - and I see they've made another post about this - is the legibility of the text, rather than the 'beauty' of the cartoons.

I think I'm right in saying that all these ereaders are 300dpi resolution, apart from the PW2. My personal recommendation would be a PW3 or a larger, maybe Kobo ereader.

PW3's are no longer sold new. I don't know where the OP lives, but they might manage to find an Amazon refurbished one. The larger Kobos are more expensive, but may be worth the money.

[j.p.s]

Originally Posted by JSWolf

I'd like you to try the following (if you don't mind). Take this image and make it a cover for an ePub. Send that ePub to the Libra with the option to dither it as grayscale. Copy the image off of the Libra and view it on each device and see how it looks. There should be no dithering from any of the devices as it's already dithered for eInk. That would show how the different screens look.

Instead, I request you use a kobo to dither NiLuJe's gray scale dither totrture test image and post that dithered image so that anyone interested can put it on their own devices for comparison. It would be nice if ilovejedd put it on hers and did another scan and/or photo test.

[NiLuJe]

@j.p.s: Err, nope, we're indeed reducing the actual amount of colors? (Given the usual non-HDR grayscale image, from 256 to 16).

Forma_Gray.png PNG 1440x1920 1440x1920+0+0 8-bit Gray 256c 414736B 0.000u 0:00.000
Forma_Dithered_o8x.png PNG 1440x1920 1440x1920+0+0 8-bit Gray 16c 91074B 0.000u 0:00.000

(i.e., reduce the *actual* amount of colors, while, ideally, making the *perceived* result not so bad. My use of 'effective' might have been confusing?).

----

Unfortunately, we can't do what you're requesting, as I mentioned elsewhere (or here, earlier): the hardware dithering is done on a work buffer that won't make it back to the framebuffer, so we can't capture it to try it elsewhere .

If you meant the dithering applied by Calibre, on the other hand, here goes.

(Again, it's easy, it's IM's 8x8 ordered dither, I'm the one who ported it everywhere I cared about ;p).

----

If you have shell access, FBInk can do that (IM's 8x8 ordered dither and/or the device's HW dithering) automatically for anything you throw at it .

[BookCat]

Originally Posted by Gametaku

Could you do a text comparison between Libra and the PWs (if its not too much trouble)?

Is there anyone with both these devices who could do this for the OP? Unfortunately, I don't have a Libra.

I think this thread has gone a little

[NiLuJe]

Text comparisons would be harder to make "objective", as the text renderer itself will be different between different brands.

Unless you compare KOReader vs. KOReader on each device, with the same settings .

But that introduces another variable that probably won't answer the original question anymore ^^

[j.p.s]

Let me rephrase.

A 256 shade gray image only shows as 16 shades on a 4 bit display. Dithering the image has the potential to effectively display 256 shades on that same 4 bit display. Effectively increasing the number of displayed shades to 256. Of course, good luck being able to distinguish 256 shades on an LCD, or just about any, display.

But your dithered example shows that dithering can help a great deal.