Does not render PDF correctly

[mbaeuerle]

Does not render PDF correctly

[mbaeuerle]

@Needles404 PR #26 fixes the issue for your PDF by using a new image overlay algorithm:

However now that I tinkered with this new algorithm I begin to understand why the standard derivation was used in the first place.
Say you have a PDF with multiple pages but the same styling, e.g. a PowerPoint presentation with a logo on each slide at the same position. The standard derivation will then strip away the recurring parts like the logo. This is most often what you want as only the differing content is what's interesting.

In your case as the pages are almost identical on purpose this approach however doesn't work very well. Maybe it therefore makes sense to offer the new algorithm as fallback if the other one does not work.

[Needles404]

Hi, yeah this issue first appeared when using 0.9, I subsequently searched for more recent versions/forks but they exhibit the same issue.

…

On Wed, 22 Jan 2020, 14:20 Carl-Erik Kopseng, ***@***.***> wrote: @Needles404 <https://github.com/Needles404> Have you tried using the original 0.9 version of Briss? It would be interesting to know if the bug is a regression or if it was present in the original source that was forked. The original Briss, awkward as it was to use, always did work perfectly, IMHO. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#18?email_source=notifications&email_token=ADZDUTIV67ABCHURDU6RG33Q7BI2DA5CNFSM4GA2Q3N2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEJTWVXY#issuecomment-577202911>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADZDUTMBS72OSU7TLKSBQJTQ7BI2DANCNFSM4GA2Q3NQ> .

[cleydyr]

I have verified this bug. The current selection boundaries are not clear when the document is loaded.

[fatso83]

@Needles404 Have you tried using the original 0.9 version of Briss? It would be interesting to know if the bug is a regression or if it was present in the original source that was forked. The original Briss, awkward as it was to use, always did work perfectly, IMHO.

[cleydyr]

I could verify that the images generated by PdfDecoder match the pages of the document. However, the BufferedImages that are being shown and which are used to produce the rectangles are already "broken". Even if I tune some parameters of the algorithms and the rectangle is shown, there's no point in cropping the document as the preview itself is "broken".

That's kinda the most I can do as I still barely understand the algorithms in all the steps that Briss uses.

[mbaeuerle]

@Needles404 I think I could track down the issue. Yeah I know, a little bit late but maybe it's helpful for you nevertheless.

As @cleydyr wrote the PdfDecoder returns a perfect image. And even if you apply cropping the resulting PDF looks perfectly fine.

It looks like the issue lies in the algorithm used to calculate the overlay image in ClusterImageData.calculateSdOfImages.
I am guessing but from the naming and looking at the algorithm I suppose the sd stands for standard derivation and basically computes how far apart each pixel value is from the mean.
Becaues every page in this particular PDF looks the same except for some parts of the barcode the black parts basically cancel out each other as the standard derivation is very small or zero.

I verified this with this prepared PDF which has 4 identical pages and the result is empty as suspected:

To fix this issue I think it is needed to exchange this algorithm. I will check how this could be done.

[Needles404]

Hey, I really appreciate your work on this, it's going to save me a massive headache. Interesting to note the reasons for the initial issue. Many thanks

…

On Wed, 22 Apr 2020, 23:25 Marian Bäuerle, ***@***.***> wrote: @Needles404 <https://github.com/Needles404> PR #26 <#26> fixes the issue for your PDF by using a new image overlay algorithm: [image: grafik] <https://user-images.githubusercontent.com/1345394/80039167-77356d00-84f7-11ea-8f55-06261a22b5b5.png> However now that I tinkered with this new algorithm I begin to understand why the *standard derivation* was used in the first place. Say you have a PDF with multiple pages but the same styling, e.g. a PowerPoint presentation with a logo on each slide at the same position. The standard derivation will then strip away the recurring parts like the logo. This is most often what you want as only the differing content is what's interesting. In your case as the pages are almost identical on purpose this approach however doesn't work very well. Maybe it therefore makes sense to offer the new algorithm as fallback if the other one does not work. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#18 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADZDUTKSQTTJTWFUKMWDCKLRN5VG7ANCNFSM4GA2Q3NQ> .

[mbaeuerle]

@Needles404 I finally had to time to finish this. You can find the new version here:
https://github.com/mbaeuerle/Briss-2.0/releases/tag/v2.0-alpha-3
Give it a shot and let me know if this works for you :)

With this new version the preview is basically falling back to another algorithm if a certain amount of the content is similar on all pages.

[Needles404]

Thanks again, it is now showing a usable preview which is excellent. The preview appears to be missing some text which I can only attribute to use of language packs since they are mainly Asian characters, this would probably be a different issue.

Package - FBA15CWS6NN8.pdf