OCR (Optical Character Recognition), Explained

What is OCR

Optical character recognition, or OCR, is the technology that turns an image containing text into digital text. A scan, a photo taken with a phone or a PDF that is really an image all contain letters a person reads effortlessly, but to a computer they are just pixels. OCR detects those letters, words and lines and returns them as text that can be searched, copied or processed.

Modern engines don't stop at individual characters. They also recognize lines, tables and the position of each word on the page, which makes it possible to know where each value sat in the original document.

How it works

In simplified form, OCR follows these steps:

• Image preparation. Orientation, contrast and skew are corrected.
• Detection. Areas with text are located and split into lines and words.
• Recognition. A model identifies each character or word and assigns a confidence level.
• Output. The text is returned together with its position on the page.

OCR for documents: necessary, but not enough

OCR is the foundation of any document automation, but on its own it doesn't understand what it reads. It can recognize "1,250.00" with complete accuracy and still not know whether that figure is the invoice total, a tax base or a line amount. Nor does it know whether the supplier exists in the ERP or whether that invoice was already recorded.

That is why, in an invoice process, OCR is an important piece, but it is not the whole process. It needs structured extraction, validations, duplicate detection, exception handling and integration with the accounting system. That combination is what is known as intelligent document processing.

Why OCR fails on some scanned documents

OCR accuracy depends heavily on the input. The most common problems are:

• low-resolution, blurry or skewed scans;
• incomplete or badly oriented pages;
• stamps, signatures or notes over the text;
• handwriting;
• complex tables, with merged cells or no lines;
• uncommon languages or typefaces.

An OCR error can also show up later, for example when the supplier's name isn't recognized because the text read doesn't exactly match the registered one. The answer is not to create a new supplier, but to combine reliable identifiers and approved rules.

A concrete example

A company receives paper invoices that it scans at reception. OCR turns each scan into text with positions. From there, another step finds the invoice number, date and total, and a third checks that the amounts add up. If the scan comes out skewed or low quality, confidence drops and the invoice goes to review instead of passing with a doubtful value.

OCR at DPI Systems

In DPI Systems' invoice digitization projects, OCR and extraction turn the document into data, and the workflow adds validations, duplicate detection, approvals and integration with the ERP.