Hello All,
I'm running a small experiment on the success rates of OCR/ICR data capture of hand printed documents.
This is really a test of human ability to successfully capture this kind of data more than a test of software capabilities.
I was wondering if anybody out there has experience in this kind of data capture?

I can't say with any degree of qualification but on the few occasions where i have used OCR i have found it to be useful but not totally reliant.

I had a document that had been printed, but the original file had been lost. Rather than retyping it i used OCR. What i found was that for the majority of the text it had no problem but for less common words or brand names it struggled. It also had a bad habit of confusing certain letters, 'r' 'n' and 'm' especially.

I found that i still had to proof-read the document to ensure it was correct. I think it was faster than retyping, but not if it had been a truely technical document, such as a dissertation or similar.

I have worked in data capture most of my working life.   Handprint ICR success factors: Forms designs and printed for ICR; business case if 75% ICR accuracy at field level for 25% effort and 50% faster results; form fillers asked to write neatly in boxes (and most do), enough volume to offset extra costs of design/print/form definition/ICR s/w/doc prep/scanning; effective manual correction function (for correction, joined up writing and ICR queries); plenty of data validation opportunities (check digits, balances, table look-ups, logic). Also may be better to use more tick boxes rather than handprint response areas, even if uses more pages. Failure factors: too low volumes, solution oversold as >90% handprint accuracy, extra print costs exceed costs/time benefits, business case cannot compete with offshore key from image bureaus, not prepared to design and print ICR friendly forms, clumsy manual correction functions, too few form fillers write neatly in boxes (or don't use uppercase). Fewer than a thousand of forms a week is unlikely to be worth cost savings, unless turnaround time is the critical benefit and if you were going to scan anyway. Lower case and cursive ICR handprint recognition gives poorer accuracy results (more choices between lower case L, one (1), stroke (/) and same for other similar characters).