After almost 2 years of maintaining tesseract.js, I have to say tesseract.js hit its limitation as some of the issues from users are hard to fix, especially those related to performance.
The existing tesseract.js (before v2.0.0-alpha) offers declarative and coarse-grained APIs that minimize the effort to understand the complex design of the library, it hides details like how tesseract.js loads *.traineddata from remote, how the *.traineddata is saved to local file system or IndexDB as cache, how the WebAssembly version of tesseract-ocr is initialed with parameters, set the image and generated the result we desired. It is great and everyone is happy to get a turn-key OCR library to integrate into your project.
But as you use tesseract.js more and more in your project, you figure out the performance is not so good. It feels like every time you call recognize(), tesseract.js always starts from scratch and loads the *.traineddata, initializes its core and set the parameters again and again. It really takes lots of time and the performance becomes unsatisfied.
Another issue is that every time you want to do something that tesseract-ocr can do with tesseract.js, it fails immediately and you totally don’t know how to solve it, so you create an issue…
