Using spellchecking to improve Tesseract OCR accuracy