Aztec recognition capability

[DrNion]

First of all thanks for the nice little app!

Since some companies use weird QRCode-ish codes which obviously cannot be recognized, it would be an awesome addition to your app to optionally cut any code (or any other part) manually out of the PDF for the preview.

Edit: ...to support Aztec Code.

[michaeltroger]

Hi!

THX for your feedback/suggestion! I experienced a similar problem with Austrian ÖBB train tickets. Their "QR code" has some symbol in the centre. Additonally it doesn't have the 3 boxes at 3 corners as the standard suggests. Anyhow it's not being recognized as QR code.

I'm using here the ZXing library for reading QR codes. ZXing is the most popular library out there. Not sure why it's not capable of detecting those variations...

Regarding cutting a part out, I must say that I would prefer a solution for the root cause. I can investigate if there is a better way to detect "special" forms of QR codes. While technically for sure possible, the cutting would be quite some effort to implement, and I'm not sure there is so much demand for such a feature that would justify the time effort.

How does your problematic "QR code" look like? Is it also one with symbol in the centre, without corner marks or what is special about it?

[michaeltroger]

Seems like there are at least 2 additional formats that look kind of similar to QR codes: https://en.wikipedia.org/wiki/Aztec_Code https://en.wikipedia.org/wiki/Data_Matrix Seems the one from ÖBB is Aztec. That explains why it's not being recognized. Does yours look similar to one of those?

[DrNion]

Yes kind of, it's one from the Deutsche Bahn. But I tried to scan their Code and my scanner didn't recognized it. I will look it up, maybe it is still a known format and nothing proprietary.

[DrNion]

Seems to be Aztec: https://stackoverflow.com/questions/34423303/need-help-decompressing-zlib-data-stored-in-aztec-barcode-deutsche-bahn-ticket There you can find a ticket example too.

[michaeltroger]

Since Aztec seems to be wildly used for public transport tickets I'll try to support it in a future version. I'll come back to you for trying it out. ZXing has some dedicated API for it

[DrNion]

Ok, I changed the Titel :)

[michaeltroger]

Hi @DrNion , sorry for the late response. I concretely tried to implement this feature, however after careful consideration, I decided to not invest in this topic any longer. I can simply not assure it works good enough on a variety of devices, without investing a lot of time.

Unfortunately it's not as easy as just searching the image for a barcode with a barcode library. In reality the image resolution is too low for detecting Aztec codes (feel free to try out the changes of the linked PR).

Now I had different solutions in mind:

• I could increase the image resolution, however I already have now issues on low-end devices that run out of memory. So this is not really an option. And my ÖBB Aztec code was not even found with higher resolution.
• I considered a new screen on which you can select the approximate area of the barcode in the document. It would allow searching in a smaller area of the document, which made the Aztec code be found in my test builds (while also increasing the resolution). However it would not only require an additional screen, but also a database update, where you would persist the location of the barcode within the document. And while not only technically challenging it's also challenging in regards of providing a meaningful user experience. After thinking it through it's a much more complex topic then I would wish for. And since it's a free time project I don't want to invest that much time.

Sorry for not having better news.

External help is welcome for implementing this though (please align with me before investing too much time).

[rugk]

I already have now issues on low-end devices that run out of memory. So this is not really an option.

Uff bad thing. However, can't you try to implement some kind "streamed processing" where you automatically "divide up" the image into squares (proportionally to the initial image size) and process each square separately? And then throw it away in memory… Oh I just see the flaw here:

• Of course this fails if the image is overlapping? -> then one would need some cool mechanism of checking each half of each pattern again…
• or limit the initial search to only the thing in the middle icon which recognizes the code?
• of course, start the search at the edges (more likely to have these codes) and maybe stop it just with some timeout (or memory usage limit)…
• or only check the positions of a darker luminescence (or how was it called in this one color model?) or some other indicator (only black and white areas? -> convert the image to grayscale and find only a strong contrast) or other such tricks maybe?
• it, maybe the simplest thing, just disable that feature on low-end devices, i.e. most modern devices have 4GB of RAM – should not that be enough?
• searching online I only found this article using tensorflow i.e. AI, likely won't help – unless we use AI to recognize the position of the code. Which, thinking about it, may not actually be a bad idea. AFAIK some phones have special processors for that and may have a thing to do that also quite with less memory-problems (I wonder how these Google photo apps or so actually work that do crazy things based on that.)… training should also be pretty simple and after all it's just about finding a position – we can then run the "normal" algorithm on that. If we find faces with AI, we should be able to find simple codes, should not we? (And Google and others AFAIK promoted AI integration into their products so you will, likely find some good libs for that.)

Just some ideas/brainstorming for some approaches… (Note I am no Android dev, so I do not know the limitations of the platform in detail.)

I see the UX challenges of the second approach, and in contrast to that I guess a few seconds of waiting for some complexer algorithm should be fine.

---

Do other scanner apps that support aztec code have the same issues? Likely not, because you as a user are expected to choose the size there, as it seems.

Btw for me being curious: Why is processing QR codes not a problem, but Aztec codes is actually?

Because that is used really really widely for public transport…

[michaeltroger]

Version 4 now supports Aztec code experimentally