Hi, I'm using Tesseract 3.01 on images basically containing two columns of multidigit numbers. The source material is semi-poor computer printouts from the 60's. I've trained Tesseract specifically for that data, using a unicharset containing only the relevant characters, and overall I'm very pleased with the accuracy. On character level, I'm getting about 99.8 percent. What I'm trying to do now is find a way to locate probable errors to make it easier to fix them.
My first approach is to make use of Tesseract's confidence data. Having researched this a bit, I realize those numbers may not do me a whole lot of good, but I'd like to at least give it a try. What I've tried so far is to patch TessBaseAPI::GetBoxtText to include a new column in the box file containing the confidence values, by calling Confidence(RIL_SYMBOL) on the ResultIterator for each character. The problem is that I get the same confidence value for all characters in a "word", rather than character-specific values. Is this what's meant to happen? I've found that for my data, best_choice->blob_choices() always returns NULL in ResultIterator::Confidence. Is this why I get word confidences, or would it be the same thing if I did get choices, and choice_it.data()->certainty() was called instead of best_choice- >certainty()? And should I be worried that there are no choices? Of course, if there's a better way of getting at the character-level confidence values, I'd appreciate any pointers you may have. Thanks in advance, Mikael -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/tesseract-ocr?hl=en.
