4 ms·
I'm not convinced OCR is as good as humans. I recent did a project making an unauthorized copy of a rare ($2000) book from a university library. I scanned in ev
by umvi 8y ago
I'm not convinced OCR is as good as humans. I recent did a project making an unauthorized copy of a rare ($2000) book from a university library. I scanned in every page, but tesseract OCR really struggled with pages that started off straight but curved off. I tried lots of preprocessing techniques with limited success. My options were to type it in by hand or rescan that page so the lines were straight.
- computerex 8y agoWhat a weird way to compare OCR to humans. If a human can't see the writing because the page surface is curving away, they'd adjust the page surface. Likewise, getting decent scans is the most cost and effort effective way of getting good performance in OCR. I personally found tesseract to be incredibly good, and have even used it in non-traditional OCR applications for doing things like reading signs.
- umvi 8y agoI'm saying that I can read skewed/bending pages easily, but I'm having a super hard time getting tesseract to play nice with such images. I almost always need to rescan. Tesseract is incredibly good... as long as your lines of words are straight.