The experimental version exists before this commit 9753889. It uses Tesseract (using pyocr). ocropy looked promising the last time I checked, opening this issue for discussion and experiments around OCR.