Tesseract is really difficult to tune such that you get decent results from camera images instead of cleanly scanned text.
Ocropus is really slow, but does seem to work better than tesseract out of the box. It's still quite bad with camera-captured images and IIRC doesn't give you any hooks to improve the results.
I couldn't find any other viable options when I was looking about six months ago.
I want an SDK that gives me an easy way to train the OCR engine. That means being able to view the images which performed badly, with text bounding boxes and the OCRed text overlaid that allow me to manually correct the results which it will then learn from. And, perhaps most importantly, good defaults for the common cases of document scans & camera captures.
Ocropus is really slow, but does seem to work better than tesseract out of the box. It's still quite bad with camera-captured images and IIRC doesn't give you any hooks to improve the results.
I couldn't find any other viable options when I was looking about six months ago.
I want an SDK that gives me an easy way to train the OCR engine. That means being able to view the images which performed badly, with text bounding boxes and the OCRed text overlaid that allow me to manually correct the results which it will then learn from. And, perhaps most importantly, good defaults for the common cases of document scans & camera captures.