A Portable QR Scanning library. Image processing is very much based on algorithms used in BoofCV. Decoding and error correction is based on zxing
Requires C++17 compiler. Seems to work with GCC and Clang.
Currently can be used in C++, Android (via JNI) and Web (via emscripten).
Demo is available here
There is an npm package
It will work out-of-the box in nodejs and will probably require special handling for webassembly file distribution in browser.
You can see pqrs-cli for example usage.
The android library is published on maven central as me.dcnick3.pqrs:pqrs-android (see https://central.sonatype.com/artifact/me.dcnick3.pqrs/pqrs-android). Add it to your project.
Then call PqrsScanner.scanGrayscale and get the scan results back. The Android API is a bit rough and is only really designed to be used as a Android CameraX analyzer. See example usage here.
I needed a QR scanning library in web application. So, I went with jsQR, which is one of zxing ports to js. What could go wrong? Apparently, a lot.
The scanner was used to scan an image projected to screen in an intensively lit big lecture room. Users were to scan it without leaving their places. And this actually posed a challenge...
I have distinguished two conditions that pose a challenge to zxing:
First of them is low projector image contrast, which is an inherent problem of all projectors. They cannot make "black" any darker than ambient lighting, so the only way to increase contrast is to make light regions even lighter, which is limited by physical abilities of the light source.
This affects the first stage of QR code scanning: detection & localization. Most scanners begin with image binarization and then try to detect finder patterns (the noticable three squares at the angles of the QR code).
Low contrast is quite a challenge for most binarization algorithms and, for no surprise, zxing struggled with it:
The second problem was due to the fact that lecture room was quite big and users were to scan QR without leaving their seats. This required scanning from quite extreme angles and zxing does not handle huge perspective distortions caused by this quite well.
With these cases in mind, I collected example images and used them to survey already existing QR scanners to find those reasonably handling my problematic cases.
That's when I found BoofCV, which boasted surprisingly good performance. And apparently, it was actually worked pretty good in these conditions. There was only one problem: it's written in Java, which is not straightforward to run in browser context.
So, I've decided to write a C++ port(-ish) of BoofCV QR scanner and compile it to webassembly using emscripten.
From this I got, on one hand, an ability to use it in much more languages than BoofCV (C++, arguably, has interoperability rate comparable to that of C) and performance (wasm is quite close to what you can get in native code).
It is not a direct port, some corners were cut (where it seemed like it would not affect performance or scanning quality), some features were added (like local thresholding when sampling QR code modules, which handled the uneven lighting).