Heavy ghosting watching comics

[Grimulkan]

Food for thought. I tried this out on my DR with some interesting results. I will post screenshots later, and perhaps the same on a Sony PRS-505, if I can get my hands on one again.

Meanwhile, if anyone feels like trying this out, I've included the test images and the experiments.

Numbered test images:
1) 1-allwhite.jpg - All white image.
2) 2-allblack.jpg - All black image.
3) 3-darktest1.jpg - Lots of white shapes/lines on a black background.
4) 4-lighttest1.jpg - Lots of black shapes/lines on a white background (perfect inversion of 3).
5) 5-text.jpg - Lots of white text on a dark background, what the DR would be used for most of the time.

For the experiments, put all 5 test images into a folder on the DR. For an image sequence of 1->3 (for example), click on image 1 in the content browser and after it loads, click on 3 in the page bar. Then close the image viewer for the next experiment. The "Please wait" ghost may be present in the first image, so ignore that for these tests.

Experiment 1: Image sequence 3->4
This changes from one image to its complete inversion. It looks like there is NO ghosting in this case.
Experiment 2: Image sequence 3->1 and 4->1
This changes from some image to an all-white image. There doesn't seem to be any ghosting here either!
Experiment 3: Image sequence 3->2 and 4->2
This changes from some image to an all-black image.

Woah! Who you gonna call™?

In the sequence 3->2, the ghost shapes are darker than the background, while in sequence 4->2, the ghost shapes are lighter than the background.

My guess is that if there is anything black in the previous image and the same spot is black in the next image, it appears whiter! Anything white in the previous image does not seem to ghost at all (I could be wrong here, I'd need to examine the camera photographs carefully to decide), and anything white in the next image does not ghost either.

So, contrary to popular belief, ghosts aren't white at all In fact, anything white is immune to haunting! You can also try the sequence 5->3, to see how ghosting could really distract. But 3->5 wouldn't produce noticeable ghosting, so normal text reading is not affected. Also, 5->1->3 will kill all the ghosts.

This is really puzzling, because I was under the impression that black and white are both equally natural states for e-ink "pixels" as they are bi-stable. Why would one color ghost and not the other? Does it have something to do with the particular refresh strategy?

I still don't know if this is a DR controller-specific issue, or a general e-ink thing.

Edit 05/20: This thread is relevant to your interests. Also see a follow up to this post here.

[Shaggy]

Originally Posted by Grimulkan

I still don't know if this is a DR controller-specific issue, or a general e-ink thing.

It's a general eInk thing. Different manufacturers build in different methods for doing the refresh sequence, but they're all effectively doing the same thing you did in your example on every refresh. So, for instance, when you go from one of your images to another, the device is also doing intermediary refreshes with white/black/inverse versions of your image (depending on how they do the refresh).

For example, on a device that automatically does an inverse step in the refresh, if you do a 3 -> 4 -> 3 sequence, what is really happening is a 3 -> 4 -> 4 -> 3 -> 3 sequence (the device is doing the "bold" steps on it's own).

If it was a device that does an automatic white step, then your 3 -> 4 -> 3 would actually be performed as 3 -> 1 -> 4 -> 1 -> 3.

...etc.

[Grimulkan]

@Shaggy: I see. Thanks for the explanation. The good news is that we can "bolster" the refresh strategy by forcing certain images to be displayed. So someone building a software viewer could enable an option to reduce ghosting, at the expense of slower slightly slower refresh.

[flowoeB]

thank you both! I did the following:

i added a blank white page between every two pages of the comic:
no more ghosting!

you have to doubletab to turn but thats ok - image quality is perfect now!
THX again!

[flowoeB]

so ghosting seams to be a pure software problem and could be corrected easyly?

[Grimulkan]

Originally Posted by flowoeB

so ghosting seams to be a pure software problem and could be corrected easyly?

No, from what I understand ghosting is inherent to e-ink, although one can 'condition' the display signal to minimize it, at the expense of other things. I suppose iRex has decided to pick the current set of refresh speeds, contrast and ghosting levels, especially because ghosting does not surface while reading plain text.

What you have done by inserting white pages (and what could theoretically be enabled automatically as a high-level software option or as low-level instructions to the delta controller), reduces ghosting but increases refresh time, something you are willing to live with in this situation

So no, there is no "correction" to this "problem" with what we have right now, but there are other things one can do to minimize it.

It would be really cool if there were low-level instructions to the controller one could use to change the tradeoff though, such as a slow refresh-low ghosting mode for comics and a fast refresh mode for text.

[Shaggy]

Originally Posted by flowoeB

so ghosting seams to be a pure software problem and could be corrected easyly?

No, it's not a software problem. But you can work around it with software, at the expense of slower page turns. The manufacturers of the device already do this, but they're optimized more for text than heavy graphics. You're seeing more ghosting than normal because of your usage pattern.

If they "corrected" the viewer so that it reduces ghosting on your comics, then the majority of users who only need the device for text will complain about the slower speeds, because for them the extra steps are unnecessary.

[flowoeB]

so my next question is: is it possible to alter settings like the turn page behaviour of the dr1000 by a little tool written by a third party developer or would we have to ask irex to add it into the firmware? like Grimulkan said it would be realy cool to adjust it to the type of document displayed - i think lots of technical docs would also benefit from slower but better display

[Shaggy]

I'm not sure, it depends on whether the page turns are done in the viewer plugins or the underlying UDS engine. Most of the viewers (except for Mobipocket) are open source, the UDS is not. I haven't looked at the code closely enough to know for sure.

Ultimately an option would be nice for those that want it, but I don't know how much demand there is. Post in the user feedback section of iRex's forum. If a lot of people request it, they may add it in to a firmware update. I haven't seen anybody else asking for it yet though.

[Grimulkan]

Originally Posted by Shaggy

I'm not sure, it depends on whether the page turns are done in the viewer plugins or the underlying UDS engine. Most of the viewers (except for Mobipocket) are open source, the UDS is not. I haven't looked at the code closely enough to know for sure.

Ultimately an option would be nice for those that want it, but I don't know how much demand there is. Post in the user feedback section of iRex's forum. If a lot of people request it, they may add it in to a firmware update. I haven't seen anybody else asking for it yet though.

As far as I know, the UDS itself is not open source, but the "inputs" to the UDS are (pdf plugin, image viewer plugin and so on). One could then modify the pre-rendering algorithms to do what FlowoeB wants.

I'm currently working on an independent rendering tool I'm writing from scratch (I simply don't think the UDS is snappy enough). It will take some time, but I'll make sure I incorporate these features, whether there is demand for it or not, because I certainly want them!

[Shaggy]

Originally Posted by Grimulkan

As far as I know, the UDS itself is not open source, but the "inputs" to the UDS are (pdf plugin, image viewer plugin and so on). One could then modify the pre-rendering algorithms to do what FlowoeB wants.

Yeah, you could do that too, IE render the extra pages. I was thinking along the lines of changing the refresh behavior, but am not sure where that code is. Either way would work.