Daemon type OCR

[marcincichocki]

When I started I left out daemon type ocr for few reasons:

• It's language dependant
• I thought ocr would be slow
• It doesn't provide many benefits

However, since other features are completed, I thought I would revisit this idea. Here are my thoughts after few hours of testing:

• Language data is available for free on github(license apache 2.0).
• Loading language takes a while(300ms) so it's a question how to distribute it and how to load it so it's responsive.
• Fragment preprocessing got big upgrade, by extracting blue channel from source I was able to clear every noise/border and other junk that would just clutter the image. This is must be implemented for daemon fragment.
• I also noticed that sharp is quite slow(150ms), I might need to see how much time does it take to process images on production.
• Text recognition is surprisingly easy, most data from registry was easy to decode, and text seems correct(can't validate for some exotic language but it looks good).
• It's super fast, and with worker support performance impact will be negligible.
• In the event of failure daemon could be marked as UNKNOWN or UNDETECTED.
• Daemon type would always be recognized to its id.
• I need list of every daemon and screenshots for for them in every language so I can be sure they work.
• I also need to know what to do when wrong ocr language is selected.
• with data in place, there has to be some use for it, as first step I could add it to the viewer, later on maybe create sequence sort based on type.

Overall there is lots to do.