OCR

AI concepts
About 1 min read

A technology that converts text in images, scanned documents, or photos into computer-editable and searchable text data.

Also known as
Optical Character RecognitionOptical Character Recognition

Detailed explanation

OCR (Optical Character Recognition) is a technology that extracts text from printed matter, handwriting, scanned documents, or photos into text data that can be processed by computers. While early methods compared character shapes with pre-defined patterns, today's technology uses deep learning combining CNNs, RNNs, and Transformers to recognize complex layouts, multiple languages, and even handwriting with high accuracy. It is used for document digitization, receipt and business card recognition, license plate reading, and converting medical and legal records. Representative solutions include Tesseract, Google Vision API, Naver CLOVA OCR, and AWS Textract.

Why it matters in tool selection

For OCR tools, what kind of input they handle well is just as important as recognition accuracy. While most tools read clean printed documents well, performance varies widely when dealing with form-heavy tables, handwriting, low-quality photos, and multilingual documents combining Korean and foreign languages. Practical criteria for selection include whether the tool extracts table structures intact, whether it provides output coordinates, and whether its throughput and unit cost fit your workload.

What to check when choosing a tool

  • Is its accuracy verified for the target document types (forms, handwriting, low-quality photos)?
  • Does it accurately recognize the required languages, including Korean?
  • Does it extract the coordinates or structure of tables and items?
  • Do the processing speed and per-document cost for bulk processing match the scale of operations?

Real-world application example

When an expense manager uploads a receipt photo, the system can be configured so that OCR extracts the merchant name, amount, and date as text to automatically input them into the accounting system. In this scenario, using a tool that also provides the locations of the amount and date fields allows building a workflow to inspect only wrongly read items, significantly reducing manual input time.

Related terms

Computer VisionCNNDeep Learning