Overdue context sensitivity in OCR (Was: Best way to scan 132 column fanfold mid-1970s text printout

On 28 Jun 2011 at 18:20, Stan Sieler wrote: Occasionally, I get a floppy in for conversion that belongs to a typewriter that I'm not familiar with. It becomes a matter of guessing which binary code belongs to what character, as many of these devices simply use the position of the character on the typewheel as its binary value. If you don't have the hardware (very often the case), you have to guess. Spaces and end-of-line sequences are fairly easy to figure out, but then the problem turns into a version of "Cryptoquote": http://www.eastoftheweb.com/cgi-bin/top_scores.pl?game=cryptoquote A bit more complicated, however, since upper- and lowercase codes are different and there is the issue of punctuation and special effects (e.g. bolding, etc.) to figure out. What never ceases to surprise me is how the letter frequency chart can vary, depending on the subject of the document used as a sample. Note on your example, where "e" is on your chart--a letter that is supposed to be the most frequently-encountered vowel and, in fact, the most frequently-encountered letter: http://en.wikipedia.org/wiki/Letter_frequency --Chuck