Text recognition works pretty well, except when there are equations or "strange" characters inside the text. A general Unicode recognition would be nice.
Fjantonio about TextGrabber scan and translate, v5.3.5