There is OCR-B[1] for that. It is widely used as the human readable part of EAN/UPC barcodes used for laser scanners in retail. The relevant thing here is that the human readable part is never OCRed in practice because the barcode can be read much more reliably. OCR-B seems to be recommended simply because it is a well specified (ISO standard), easy to use (no weird licensing), high legibility font. Which is interesting because, as mentioned elsewhere on this thread, you don't really need a special font for OCR anymore. So it is a commonly used font that no longer serves the original purpose of the design. If you trained your OCR only on OCR-B you would likely get get some amount of accuracy improvement, but that would work for any high legibility font.

So the problem would be convincing app designers to read both things, where one of those things is much more reliable. I guess that might make sense for things that require high security, where the false negatives were worth the bother.

Getting back to the original article, OCR-B only has the good OCR characteristics for the upper case Latin characters, like with the QR code size thing, and for the same reason. If you are identifying something you generally want to use the larger, easier to read (both for people and machines), upper case characters. The lower case glyphs were added later to OCR-B as an afterthought.

[1] https://en.wikipedia.org/wiki/OCR-B

Taking a QR code that takes you to a URL, and writing out the URL in text does not enable anything, as you still must have an internet-connected device to retrieve the content, in which case you would most likely be able to scan the QR code anyway.

The point is to have e.g. a QR code, while still having human-readable content for those not having or wishing to use their smartphone in this interaction. E.g., just having the list of things on the menu printed in plain text (e.g., on a wall, over the counter, paper menu, ...), but also having a QR code with images and the ability to order directly to your table - stuff beyond what text would get you.

I often see things that I want to "look up later" while I'm passing by, but they only have QR code. I don't know if it is worth the time to stop and scan but if there's a URL I can just read it and remember.

"Easy to remember URLs" are just domains that mostly match the human-friendly name of the place. As you are not memorizing a full URL, you can only really go to the frontpage, and so you can just memorize the name of the place instead.

It's a different story if there's just a random QR code with no clear purpose of ownership, but... maybe don't scan that.

Scenario 3 the poster says something is happening like "Neighbourhood dinner, Sunday at 7 -- Want to help cooking? Scan this code" -- There is a QR code but no information about who is organising it, where it is or how else to contact someone about helping to cook.

Yes, a name serve just as well as a URL, the point is that people begin to believe that a QR code is more convenient than text. Give me a link, give me a name, give me a search term, just give me something more than a QR code.

So why are physical venues in 2025 presenting me with QR codes? If there is some randomish number (like another commented pointed out QRs may have a UUID number) or a checksum, then encode those randomish bits as a dotted rectangular outline or a line underneath, so a big QR doesn't ruin my human experience.

I don't know why the parent is getting downvoted...they bring a up very good point that hidden inside these QR codes are a bunch of tracking bits and the QR's ability to include tracking stuff may sadly be a significant reason why we see QRs way too much.

Scanning a multi-page whole menu with a camera does not make it machine readable. And frankly, the menu is not in the qr-code, it is a link to the menu. What sounds appropriate here is to write also the content of the qr-code with plain letters, so one can type it, alongside the qr-code, for those who do not have a camera or whatever.

Now, the problem of requiring a mobile device to see the menu is a problem on its own, and while this is faciliated by having a qr-code vs having customers manually copy links on their phone (either writing them or with OCR), it is 2 separate issues for discussion. Moreover, OCR-A is not needed for any of that anywhere in the 2020s that we live.

The problems with QR code is that sometimes people have older smartphones or camera do not recognise them or, frankly, it being a means of obfuscating sth from humans. I have been in many situations with queues of people struggling for indeterminate reasons to scan a qr code to go fill up sth. If there was a simple link in plain text people could have even shared it between each other. Blindly trusting technology that can easily fail and is unnecessary for what one does and without redundancies is unwarranted imo.

no.

It is not the answer, it is a frustration where you wonder what "bean massacre pastry" is (chopped nut cookies, aka slivered almond cookies) or what they mean by "Surprise coriander special" described as "flavor of comatose with many spices in hot cow" as the translation. The accurate translation would be "mixed spice beef special" and "Beef with spice and vegetables."

Cameras are betterthan they were 10 years ago but machines are better when they have real sources in front of them

In my experience, there is usually an obvious difference between seeing a phone used as a camera versus not, based on whether it's aimed at an interesting subject or not. There are exceptions, like sitting with your elbow propped on the arm of a chair for a few minutes to avoid fatigue, which causes the phone to be at eye level and therefore perfectly vertical, but this is rare.

Getting hung up on the fact that the printed code needs some height to work isn't productive to the conversation.

   discord.gg/{... a few random characters ...}

My own 'discovery' about QR codes a few years is that you can make them "module 2" sized that ought to be easy to read with a low-spec system and have astronomical capacity if you use uppercase characters, a reasonably short domain and identifiers similar to random UUIDs. These were part of the system of "three-sided cards"

https://mastodon.social/@UP8/111013706271196029

but new-style cards put the QR code in front because (1) I have a huge amount of glossy paper that I can't print on the back of, (2) you can't read the QR code on the back if the card is stuck to the wall with mounting putting, (3) three-sided cards struggled with branding in that people didn't really understand the affordances they offered, a problem that the new-style cards attack in various ways.

https://mastodon.social/@UP8/113541119391897096

(Note the QR codes on both of those cards do not point at safebooru but at a redirect that I control that fits my QRL specification)

Personally I don't think any QR code for the web should ever require more than a "module 2" QR code and that printing a QR code which requires extra alignment markers is a sign of failure. (e.g. sure you can make a QR-code with 2000 bytes of form data embedded in it, but should you? Random UUIDs are so numerous and redirects so cheap that every new-style card like that Yakumo Ran card has a unique id because with inkjet printing it doesn't cost anything more)

> For our code, encoding mode is Alphanumeric (2), but we haven't implemented how to read that yet. Sorry! Try another QR code!

The original "test blanks" for UPCs were steel plates with machined cuts in them, photoreduced down to small sizes for testing.

tl;dr: It is sadly not the most efficient encoding (and they missed an opportunity to make it actually base41, which could have been URL safe) -- as defined it only needs 41 characters (as 41^3 > 2^16).

The RFC is also not standards track, it's just "Category: Informational".

I think a better approach is to understand there are many circumstances where different sets of characters make sense for encoding data. There's no need to write an RFC, instead define a custom alphabet for them, using something like base-x[1].

[1]: https://github.com/cryptocoinjs/base-x

[1] 32 original bits = 4 original bytes = 6 base45-encoded letters = 33 bits in the alphanumeric mode, so the overhead is 1 - 33/32 = 0.03125 for 4n bytes of data.

But don't you want 10^x to be slightly bigger than 256^y, so you could represent all length-y byte sequences in x-digit number? In this direction, there's 10^53 > 256^22, but that is still in bignum land.

And while 3% overhead is fine, encoding decimal digits in groups of 3 is only 0.3% plus a couple bits to round up to the nearest digit.

Also, when I do the math alphanumeric is the most efficient QR mode, although just barely.

And base45 is less efficient than looking at the efficiency of raw alphanumeric.

Not according to my math:

Numeric: 1000/1024 = 98%

Alphanum: 2025/2048 = 99%

Byte: 191/256 = 75%

Kanji: 13/16 = 81%*

Alphanumeric is the most efficient QR code encoding mode.

(Just to further make this clear, for QR Byte encoding uses ISO/IEC 8859-1, where 65 characters are undefined, so 191/256, which is ~75%. If character encoding isn't an issue, than byte encoding is the most efficient, 256/256, 100%, but that's a very rare edge case. Also, last time I did the math on Kanji it was about 81% efficient. *I have not dug too deep into Kanji and there may be a way to make it more efficient than I'm aware of. I've never considered it useful for my applications so I have not looked.)

That is a semi-correct calculation of the wrong number. Base45 does not use all 45 characters in every slot. It goes 16 bits at a time, so the character storing the upper bits only has 2^16/45^2 = 33 possible values.

The most straightforward way to measure efficiency is to see that base45 takes 32 source bits, and encodes them into 33 bits. The way you're calculating, that's only 50%

But the better way to calculate efficiency is to take the log of everything (in other words, count how many bits are needed). Numeric is log(1000)/log(1024) which is 99.7%. Alphanum is 99.9%. Base45 is 97%.

And I don't know where that kanji number came from. It stores 13 bits at a time, mapping to 8192 shift-JIS code points, and the vast majority of them are valid. It's pretty efficient.

Huh? I don't necessarily care about an exact "base45", I care about QR code alphanumeric, which just so happens to be a (generic) base 45 character set. For QR code, two characters are encoded into 11 bits.

>in every slot.

I've worked with the QR code standards pretty seriously and I am unfamiliar with the term "slots" being used by the standards. This is why I suspect your referring specifically to RFC base45 (although the term isn't used there either), which QR code doesn't care about. I also don't care about RFC Base 45 and would prefer to use a more bit space efficient method, such as using the iterative divide by radix method, which I also call "natural base conversion".

> base45 takes 32 source bits For QR code alphanumeric, 6 characters use 33 bits, not 32. way to calculate efficiency

The way we calculate this, for example, 2025/2048, we've termed "bit space efficiency". I'm not sure how commonly adopted this term is used in the rest of the industry. On the matter, I thought I had read "the iterative divide by radix algorithm" in industry, but after searching it turns out to be a term novel to our work.

This is also similar to the way Shannon originally calculated entropy and appears to be a fundamental representation of information. Of course log is useful, but it often results in partial bits or rounding, 5.5 in the case of alphanumeric, which is somewhat absurd considering that the bit is the quantum of information, again as shown by Shannon. There is no such thing as a partial bit that can be communicated, since information is fundamental to communication, so the fractional representation we've found to be more informative and easier to work with.

Granted, in all of this, when I have done the math (and I done a lot of math on this particular issue) there appeared to be some very extreme edge cases at the end result of the QR code where some arbitrary data encoded into QR numeric was slightly more efficient than alphanumeric, but overall alphanumeric was more efficient almost all the time. There are other considerations, like padding and escaping, that makes exact calculation more difficult than it's worth. I just needed to "most of the time" calculation and that's where I stopped.

For more detail of my work, my BASE45 predates the RFC by 2 years in 2019, then I published a base 45 alphabet, BASE45, by March 1, 2020, a whole year before the RFC. A patent including BASE45 was submitted June 22, 2021: https://image-ppubs.uspto.gov/dirsearch-public/print/downloa...

Matter of fact, because of the issues and confusion surrounding base conversion, I wrote this tool in 2019:

https://convert.zamicol.com

It is the first arbitrary base conversion tool on the web. It also was essential for our work with QR code and other base conversion issues.

> I suspect your referring specifically to RFC base45

> For more detail of my work, my BASE45 predates the RFC by 2 years in 2019

The RFC was linked in the comment I originally replied to. The same comment where you saw the term "base45", because I didn't repeat it in my original reply.

> The way we calculate this, for example, 2025/2048, we've termed "bit space efficiency". I'm not sure how commonly adopted this term is used in the rest of the industry.

It's not a good metric when the size can vary.

3/4 uses 75% of the bit space, and 512/1024 uses 50% of the bit space. But if you give 20 bits to each, the first method can encode 59049 combinations and the second method can encode 262144 combinations.

> which is somewhat absurd considering that the bit is the quantum of information, again as shown by Shannon. There is no such thing as a partial bit that can be communicated, since information is fundamental to communication, so the fractional representation we've found to be more informative and easier to work with.

You can use any base and the math is roughly the same.

Distinguishing between two symbols is just the minimum. You can't transmit .3 bits but you can easily transmit 2.3 bits. If your receiver can distinguish between 5 symbols at full speed then 2.3 bits at a time is the most natural communication method.

> There are other considerations, like padding and escaping, that makes exact calculation more difficult than it's worth. I just needed to "most of the time" calculation and that's where I stopped.

Yeah, that's fine. They're both efficient. My deciding factor is not the tiny difference in efficiency, it's the ill-behaved symbols in alphanumeric.

However, in such cases, they usually use only capital letters. In Japanese, there's no distinction between lowercase and uppercase letters. So for them this distinction is, if you allow some leeway, similar to the differences between normal, italic, or bold letters in English. So it makes sense with this context, if you are making a "group of letters and numbers", to default to uppercase + numbers as the normal/shorter version.

I decided to use segments and stick to lowercase "https", because I didn't trust various implementations out there to handle "HTTPS" correctly. Should I?

RFC 3986 explicitly says schemes can have uppercase letters but are canonically lowercase, and says that they are case-insensitive. But it also still says that implementations "should" accept uppercase letters as equivalent.

WHATWG's URL standard mandates case-insensitive matching.

In practice anything you scan a QR code with will probably not have a problem with the uppercase scheme.

Although we found browsers were out of alignment with standards on all sorts of matters, we found broad compatibility with upper case. (Of course, meaning everything before the path. The interpretation of the path is delegated to the server which may or may not be case sensitive, up until octothorpe, #, which is then solely interpreted by the browser.)

Leaving a naked domain name like "www.example.com/1234" was not quite as good, but at least iPhones, Pixels and Samsungs worked well IIRC.

I used this trick to allow printing of QR codes at smaller resolutions for 4+ years, and phones have gotten noticeably better in that span at handling QR codes printed at smaller sizes, with uppercase, nonstandard shapes, borders, you name it.

If you make https: lowercase and the rest uppercase, with QR encoding that does smart segmentation, I'm pretty sure you can still get most of the benefit... but, exercise for the reader.

The smallest v1 QR code is 21 modules. That can fit 20 uppercase letters.

The 25 module size, is not that much bigger and can get 38 uppercase or 26 lowercase letters.

That is actually exactly what I ended up doing. I care about all mobile phones and tablets, and I was worried whether any implementers actually tested uppercase protocol names.

In the old days, Yahoo's build of Apache (yApache) included an option to automatically lower case urls before matching them. Super handy because lots of urls were coming in from print publications and you could never get publications to show urls properly, nor get users to type them in properly.

8 bits is enough to represent the entire ascii char table, there must be some other limitation going on. QR code control chars maybe?

The linked "byte mode" table only has 45 individual chars. This could be represented with 6 bits with room to spare..

The specified capacity of "25 characters" for QR code size is 25 characters in alphanumeric mode, not in byte mode.

> The linked "byte mode" table only has 45 individual chars. This could be represented with 6 bits with room to spare..

Even better than that - it's 5.5 bits per character! Each pair of characters is represented as a single 11-bit code unit. (This works because 45 x 45 = 2025, which is just barely under 2^11 = 2048.)

There's apparently some support in the QR standard for mixed-encoding codes, but few encoders seem to use that.

No, that link is for alphanumeric mode, which uses 5.5 bits per character (45 * 45 = 2025 <= 2048, so it fits in 11 bits).