deloptes <delop...@gmail.com> writes: > Joe Pfeiffer wrote: > >> I used tesseract-ocr, mentioned previously, a couple of years ago with >> very good success. Also, the problem he's trying to solve is much > > what means very good success? You had to proof read it at the end - time > spent. For me either something works or it doesn't none of them worked even > close to good
Yes, I did have to proofread, and of course that took time. But if you're setting perfection as the only acceptable performance level, no OCR software will ever achieve it. Human eyes don't meet that standard. I wasn't keeping track of statistics (I wasn't conducting an experiment, I had a pamphlet that needed to be recreated and then edited), but the results were very very close to 100% I certainly spent a lot more time on reformatting and editing than I did proofreading. >> simpler than the general OCR problem; he's got the actual correct pixels >> (rather than a scan), and maybe even have knowledge of what fonts are >> used. That makes a huge difference. >> > > I doubt it - really! Let me know at the end. I am curious. > > regards
