Southern Weather Discovery

Vimeo freezes accounts after malware hunts for logins, coronavirus map app infected with evil code, and more

https://twitter.com/i/web/status/1239779385999159296

IIRC, there are only 3 thermometers used to calculate (Make up) the national average temperature for NZ. One on mainland (South Island) and two on the north island, Wellington and Dorkland.

Working in various departments at a government Department known as the United States Postal Service, one learns some daft things from an engineering department that was ranked in the top global experts for Optical Character Recognition (OCR).

Still, every few years, some lower level manager would dream up a wonderful “What If” solution to their archaic methods of records (data) collection, retention, recall, etc.
Those imaginary scenarios always involve Information Technology devising some miraculous OCR solution for non-IT departments, that eliminates people. Often without involving those tiresome engineers who insist on silly things like standards and definitions.

None do.

Let’s start with the basics:
1) Paper with ink, pencil marks, stains form the basis for written communication and records collection.
Ink alone comprises an astounding array or products meant to leave often colorful marks.

Writing with ink starts with fresh ink and ranges through declining marking ability as the ink runs out to a final often discontinuous vague markings.

Pencil leads are a wide range of compounds all intended to allow a substance, e.g. paper, to accept and retain some eroded lead/charcoal.
Again, depending upon the compound, that marking ranges from bold thick lead layers through slight shades of harder and often older leads.
There are substantial reasons people are told to use pencils with 2B lead for filling out important papers like income tax forms, college exams and election vote submissions.

All of this comes before the near infinite myriad of forms people devise to record information upon or the common practices of the day in writing/print styles, error corrections, symbols designation.

A quick example are all typewritten communications prior to the 1990s where the number 1, the small l and the capital I were the same. All to save complexity in the typewriter or print devices of the day.

Ignoring the issues surrounding imaging written or printed information upon documents (Contrast, Reflectance, Resolution, etc.) one must deal with the issues surrounding basic communication detail itself.
Is that a Capital I?
or the number 1?
Or perhaps a lower case L?

Frankly, it takes a human brain to note all of the surrounding context to guess the information intent. Even then, it is almost impossible to decode documents where the capital I, number 1 and lowercase l are the same.

2) Writing conventions are very different even within small time frames.
Is that a french 7?
Is that a french Z?
Both of which have Nothing to do with the language or culture of France; instead it is a horizontal line through the supporting vertical slanted line. Supposedly to definitively separate 7s from 1s or Zs from 2s…

Is that a 9 or a written g?
Is that a 6 or a sloppy C or G?

3) Data recording forms are wildly different in size, paper composition, design, format, columns and line entries.

Returning to the USPS; where addressing standards are well known and practised for decades; where paper composition and size are defined by regulations; where letter fonts and size are defined by regulations.
Handwritten mail are still conundrums. Frequently requiring humans to decipher.

The Post Office Operates Remote Encoding Centers (REC) where images of mail that that OCR failed to interpret are sent to queues that are fed to people via computers and screens.

As a typical government operation, the Data Conversion Operators, i.e. employees, are tasked with strict and often optimistic production standards. Meaning that employees are pressured to complete and input addresses.

All errors in OCRing addresses end up in customer mailboxes. Often accruing cross outs and additional writing as customers and USPS employees try to interpret who really is supposed to receive a letter.

USPS engineers and IT employees were actually quite clever back in the 1990s. OCR reading, even of well defined bar code markings are utterly dependent upon reflectance and contrast. What one machine reads under certain conditions another machine may fail to read.
USPS engineers and IT workers devised a method that when a mail piece is successively OCRd, they print a bar code on the back side of the envelope using ultraviolet activate ink.

Printed black on white addresses and bar codes, under the best conditions are successively usually read successfully at less than 95%. UV activated bar codes are always successfully read at better than 98%.

Anyone who has tried to read and decode old handwritten documents can attest to the difficulty.
e.g. Census forms:
Every Census used different forms. Frequently there are regional differences.
Every Census form was handwritten, frequently very sloppily.

In trying to decode information about my ancestors, I have flipped through many pages of Census forms filled in by the same Census taker trying to definitely decode their writing.
Fancy cursive writing styles sloppily written can be nearly opaque to understanding.

What every OCR attempt often results in is that all results must be carefully assessed for errors and corrected; repeatedly.
A change in columns that rearranged where certain data are entered may never be identified.

Good luck Niwa…