GlossaryKind of term: Manufacturing termOCR (optical character recognition)
What is OCR?
OCR (optical character recognition) converts characters in an image, such as a scanned drawing, into text that software can process further.
Letters, digits and symbols are all in scope. On an engineering drawing that means dimensions, tolerances, fits, thread designations, the material in the title block and the notes. OCR delivers characters, not meaning: that ⌀12 H7 is a bore with a tolerance class only follows from interpreting the drawing. A scanned or photographed drawing is an image and needs OCR.
What does OCR depend on?
Whether OCR is needed depends on the file. A PDF straight from the CAD system often already holds its text as text, which can be taken over without recognition, although symbols from special fonts sometimes come through as other characters. Some exports turn text into lines, and then even a digital PDF is an image for this purpose. With a scan, the result depends on resolution, contrast, skew and how many copy generations it has been through. Small and stacked characters are the most error-prone: the limit deviations of a tolerance, the tolerance classes in a fit, the pitch of a thread, a decimal separator.
How does OCR work in practice?
A wrongly converted character changes a requirement: a 1 that comes out as a 7, a lost decimal point or a ⌀ that comes out as a 0 can change a dimension or tolerance, and with it the operation and the price. An OCR result is therefore a proposal until someone has confirmed it against the drawing. Given a poor scan, ask first for the original PDF, or the model with PMI.
What is OCR often confused with?
OCR is not drawing reading: OCR converts characters, drawing reading interprets them together with the views, datums and standards the drawing refers to. OCR is not the opposite of computer vision either; it is a branch of it, and computer vision also covers finding symbols, frames and views.
Frequently asked questions
Do I need OCR for a PDF from the CAD system?
Usually not: such a PDF often already holds its text as text. OCR is needed for a scan, a photo or a PDF in which the text was written out as lines. Check symbols separately; they sometimes come through as other characters.
Which characters are most error-prone in OCR?
Small and look-alike characters: 1 and 7, 0 and O or ⌀, 5 and S, a comma or point, and the stacked limit deviations of a tolerance. Those often decide the requirement, so check them against the drawing.