Could you consider develop a feature that showw the original text of the annotated area? A user may choose to switch between seeing the text layer, or the captured text. The problem is because highlighting math formulae usually results in nonsense characters.