Optical Character Recognition (OCR) forms the bridge between the analog world of paper and the world of machines. The modern-day expectation is that when we point a smartphone camera at some characters it will flawlessly recognize and read them, but OCR technology predates such consumer technology by a considerable amount, with IBM producing OCR systems as early as the 1950s. In a 1960s promotional video on the always delightful Periscope Film channel on YouTube we can get an idea of how this worked back then, in particular the challenge of variable quality input.
What drove OCR was the need to process more paper-based data faster, as the amount of such data increased and computers got more capable. This led to the design of paper forms that made the recognition much easier, as can still be seen today on for example tax forms and on archaic paper payment methods like checks in countries that still use it. This means a paper form optimized for reflectivity, with clearly designated sections and lines, thus limiting the variability of the input forms to be OCR-ed. After that it’s just a matter of writing with clear block letters into the marked boxes, or using a typewriter with a nice fresh ink ribbon.
These days optical scanners are a lot more capable, of course, making many of such considerations no longer as relevant, even if human handwriting remains a challenge for OCR and human brains alike.
I still remember the first time I saw OCR system. It was at JCPenny store, they had a device resembling a small water gun to scan the tag on clothes or merchandises. They had to swipe across the code for the system to read and return the price amount. The font used was a very specific font to make OCR easy and still readable by human for the dollar amount
Tangentially related, the drive towards “paperless office” resulted in literal/uncontrollable explosion of paperwork. While some of it is truly paperless, the rest is not, and as any office, commercial of government, will tell you now, they are permanently short-staffed, and the data entry positions (the ones that pay absolute below minimum wages, too, btw) are where the largest turnover is.
I worked or a gate array chip that performed the OCR operation. Each chip was looking to match 1 character in 1 font. 31 chips on a board and many boards in a cabinet. Time frame was when credit cards were being run through that roller mechanism. Most interesting feature was after the electronics did it’s work all questionable results were passed through a spell checking program before shown to a human.
OCR technology used primarily in the scanning of cheques in banking.
The need for verification of account numbers, value of the transaction, removed the need for human intervention
I think most checks, at least in my lifetime, have used MICR which is a little different (magnetic ink).
Came here to say this. So thank you.
I worked in a lockbox for years. No, not Al Gore’s lockbox. It was a term used in the industry. Remember way back in prehistoric times when bills would come in the mail and we sent back payment with a pre-addressed envelope (postage required) with a remittance stub and a check (no cash please)? Well those payments went some place. Banks and third parties referred to those places as ‘lockboxes’.
Many, many thousands of mail processed every hour 24×7 and almost 365 days a year even on holidays. Multiple runs every day from major USPS distribution centers in trucks and vans. No mail deliveries due to a holiday. We still got the usual mail deliveries. Banks not open on a holiday? We still made multi million dollar deposits throughout the day. Every hour of interest counted!
The equipment was impressive. Some of it bleeding edge tech for the day.
The remittance stubs were pure OCR as described. A long string of digits containing the account number, monthly billed amount and total balance due if previously not paid. Zeros for padding and a checksum at the end. Any misread or unreadable strings required some manual entry or completely kicking out to manually research. The initial machines I worked on were using early 80’s tech and each machine could handle around 10,000 transactions per shift mostly limited by human factors.
After that hybrid manual/machine processing with the checks endorsed with a rubber stamp mounted on an aluminum cylinder (remember to change the stamp for each biller!!!) the checks went to large sorting machines. Those check sorters read the MICR Routing and Transit digits. No real OCR. Pure MICR. Very, very fast and very accurate. The sorting was done to break out major banks first and then group others primarily by region. Every level of additional sorting we could do meant less the banking system needed to do which gave us cost breaks. Even fractions of a penny added up quickly with the volumes involved. We only had two of the sorting machines (nicknamed Freddy and Jason with life sized cutouts of the movie characters by each) since they cost about $250K each in 1980’s dollars. But they paid for themselves. Just worth a ton of money and got lots of love and care. Even when the new roofing leaked, we rushed to cover those two up with tarps lest we lose them and be down for weeks to get new ones produced and installed!
Why not use MICR everything? Cost. Magnetically impregnated ink is more expensive than plain old ink. Again, at volume it adds up. OCR was just fine for remittance stubs. MICR was a banking industry standard so why not take advantage of it.
Hand writing recognition for payment processing was bleeding edge at the time I worked in that business. Being electronics and programming aware, I got involved in everything from repairing my own workstations to fixing other’s workstations and then testing and troubleshooting the new tech making it’s way in to the industry.
One funny piece of equipment was a Change of Address checker! Oh boy! So back in those prehistoric times, remittance stubs usually had a box to check if you moved and then some fields usually on the back to put the new address. Manually, many employees missed those or simply didn’t care. So some small company came up with a contraption to auto detect COAs. I do mean a contraption. Wish I had a picture of it! Start with a money counter as seen at banks. Instead of cash, remittance stubs were handled one at a time. Very fast! But that meant the detector needed to be very fast too! Now consider late 80’s/early 90’s video tech on PCs. So this PC ran Windows and had a one-off program written in C# if I recall correctly. I got past the rudimentary security to troubleshoot and fix the software when the vendor went MIA. The hardware around that cash counter was the star though! Multiple B&W video cameras that were pricey. A video capture card that alone cost a reported $10K. And for lighting? Oh boy. Ran best in a darkened room to allow the multiple strobe lights to really capture with enough contrast at high speed if the box was checked and there was writing on the address fields on the back. Not for use by those with epileptic tendencies. Quickly gave everyone headaches regardless.
For handwriting recognition, it was just as difficult but in other ways. Just trying to read check amounts in the numeric box let alone the written out section at high speed. Ugh. I got that $100k prototype working though. Had to MacGyver some parts in the hardware column. Made recommendations for design improvements for some parts to NCR who was trying to develop the technology. And the head software developer was a hot mess personally and professionally. But on the professional side, she had zero version control discipline. So once again had to get around the security (this one ran on OS/2 Warp) and then troubleshoot then resolve her code. I dreaded anytime she showed up in town to install ‘fixes’. Usually meant a few days wasted of my trying to undue old bugs she reintroduced and uncover the new ones. But we at least proved out the tech was usable. The sales rep LOVED us!
I look at handwriting recognition today and just shake my head. While it is not 100%, it is so far an improvement over the early years even in my limited experience.
Is that still true after CHECK 21 came into effect? I thought true micr was basically optional now that optical scanability is required.