I'm using pdf.js and RecogitoJS to build a PDF annotation interface. This seems to work fairly ok. Except there's a small offset between the rendered canvas image and the hidden text layer that's used for selection underneath.

It's probably less than 2 chars in all cases, but it still leads users to make incorrect selections in many cases. I assume browser characteristics, system font size, as well as fonts used in the PDF may play a role, and perfect alignment is impossible. But if anyone has experienced this, and can give advice on tweaking the alignment between canvas and image layer, do let me know.