OCR

OCR is the technology that made the digital archive searchable. A scanned image of a page is just a picture — a grid of pixels that a computer sees the same way it sees a photograph of a tree or a face. Without further processing, that image cannot be searched, indexed, or read aloud. OCR transforms the image into text by recognizing the shapes of individual characters and translating them into machine-readable code: the letter "a," the letter "b," the space between words, the period at the end of a sentence. Once a page has been OCR-processed, you can search it, copy from it, translate it, and have it read aloud by a screen reader.

The scale of what OCR has made possible is difficult to fully appreciate. The Google Books project, which has digitized approximately 40 million volumes, applied OCR to each scanned page, making the full text of an enormous portion of human publishing history searchable for the first time. The Internet Archive's Open Library does something similar with public domain works. Millions of books that were physically inaccessible — in libraries far from you, in conditions too fragile to handle, in languages you'd need to travel to read — became searchable texts. This is one of the genuinely transformative things the internet did to knowledge, and OCR made it possible.

The accuracy of OCR varies significantly by input quality. A clean, high-resolution scan of a modern book with a standard typeface will typically achieve 99% or better accuracy — meaning roughly one error per hundred characters, which is barely perceptible in a full text. An old, foxed, low-resolution scan of a 19th-century book in an ornate typeface might achieve 80% accuracy or worse, which produces text full of misrecognized characters (the classic "rn" misread as "m," the "l" misread as "1," the "u" misread as "n"). At low accuracy levels, OCR text is still useful for keyword searching but unreliable for close reading.

For personal library management, OCR is most relevant in two contexts. First, book scanning apps that extract metadata from photos of book covers or spines use OCR to read the title and author text — this is how spine-reading shelf scan technology works. Second, anyone creating digital copies of their own books (for archival or accessibility purposes) will encounter the question of whether to run OCR on the scans. A PDF without OCR is a collection of images; a PDF with OCR is a searchable document. For any text you might want to search or have read aloud, OCR is worth running.

The limitation of OCR that matters most for book culture is also its most interesting implication: OCR recognizes characters but not meaning. It can tell you that a page contains the characters H-a-m-l-e-t but not whether those characters appear in a scholarly note, a character list, or the famous soliloquy. Context, nuance, and the visual communication of layout — all invisible to OCR. The searchable text it produces is a kind of index: it tells you something is there, but the reading still requires human attention.