Grades 6–8 · History & Evidence
Why can OCR text differ from the scanned page?
A scanned page is an image of visible marks. Optical character recognition, usually shortened to OCR, tries to identify letters and numbers in that image and turn them into machine-readable text. The image and the text are therefore related representations, not the same object.
In the diagramAn image and OCR text are not identicalOriginal drawn scan model · no real archival pageSource imageTotal 18Fainter marksRaw OCR textTotal 1BMachine transcriptionThe image has 18; OCR has 1B.
OCR is useful because software can search character strings much more easily than it can search page shapes. But faint ink, unusual letterforms, damage and complex layouts can lead to recognition errors. A missing search result may reflect an error in the extracted text rather than absence of a word on the page.
Our drawn source line reads “Total 18.” In the initial machine transcription, the final 8 becomes a B, producing “Total 1B.” The visible image has not changed, but searching the extracted text for 18 could miss this line. Copying the OCR string would also carry the error into a quotation.
Review compares the extracted text with the source image. In this small model, that comparison restores 18 while leaving the image pixels alone. A faithful transcription still does not prove that the original author’s total was factually correct; recognition accuracy and historical accuracy are different questions.
About this explanation
Starting knowledge: Distinguish an image of writing from editable characters and recognize that software can misidentify a mark.
Sources
- Library of Congress — Chronicling America: Technical Information
- National Archives — Transcription Tips