I use the hocr feature of tesseract and parse the resulting xml that contains all the text with bounding boxes.