In more detail
An OCR engine first cleans the image (converting to black and white, straightening skewed lines), then finds blocks, lines and words, recognises each character and uses a language model to pick likely words. Accuracy depends on image quality: a resolution around 300 dpi, high contrast, straight lines and common printed fonts give the best results; handwriting, decorative fonts and complex tables are much harder.
Why it matters
A scanned PDF or a photo of a page is only a picture: you cannot search, copy or have it read aloud until OCR adds a text layer. Always proofread the output, since similar characters such as 0 and O, 1, l and I, or rn and m are easily confused, and a wrong digit in an amount or account number is worse than an obvious error. Setting the correct language improves accented characters noticeably.
Example
An A4 page scanned at 300 dpi is about 2480 × 3508 pixels, a good size for OCR. With the open-source Tesseract engine:
tesseract scan.png output -l eng
This writes the recognised text to output.txt; -l spa or -l fra selects Spanish or French.