BURGER & KABAB HUT INC

General technology reading / 04

Turning a picture of text into text

Illustrative line study

A photograph contains the appearance of letters, rather than searchable, editable characters. Optical character recognition, or OCR, attempts to turn those visible shapes into machine-readable text.

Preparing the image

Recognition can begin with adjustments to the image, such as correcting a tilted scan or separating text from background marks. Software then identifies likely characters or words. Layout analysis may also help determine which pieces belong together and how the page is arranged.

Recognition needs review

A mark that looks clear in context can still be ambiguous to software. Small print, damaged pages or unusual lettering can introduce mistakes. Keeping the original image alongside the extracted text makes comparison possible. Searchable text is useful, but its existence alone does not establish that every character is correct.

An illustrative example

Suppose a hypothetical community archive scans a printed catalogue. OCR could make names searchable, yet an incorrectly recognized name might disappear from search results. Reviewing important names against the image would address a different problem from simply making the scan sharper. A useful question is therefore what the extracted text will be used for, and which errors would matter most in that use.

Further reading: IBM’s OCR overview.