Vector vs. Raster Engineering Drawings: Differences That Matter

Two kinds of PDF that look the same on screen

A PDF is a container. The same drawing sheet can reach you as drawing instructions or as a photograph of paper, and at normal zoom the two look alike. The difference decides how much of the drawing a person or a program can read without retyping it, and how much of what is read can be trusted.

Native vector PDF

A vector PDF is usually written by the CAD system itself, through an export command or a PDF printer driver. Each visible line is a path: straight segments and curves with coordinates, a stroke width and a dash pattern. Dimension values, notes and title-block fields are text objects: characters in a font, at a position, with a size and a rotation. Zoom to any level and the lines stay sharp, because the viewer redraws them from the stored geometry.

Two details matter for anyone extracting data. PDF has no circle or arc primitive, so a hole outline arrives as cubic Bézier curves approximating the circle (often four of them, sometimes many short straight segments), and software has to recognize those curves as a circle. Nothing in the file says that a given text string belongs to a given dimension line either. The association that existed in the CAD drawing is usually gone, leaving lines and text that happen to sit near each other.

Scanned or raster PDF

A raster PDF holds an image: a grid of pixels for each page, from a scanner, a copier, a photograph or a print-as-image setting. There are no lines and no characters inside it, only dark and light pixels. Software has to find the characters (optical character recognition, OCR) and the lines (vectorization) before it can read anything, and both steps produce guesses with a chance of error.

Hybrid files: searchable scans and text exported as strokes

Two hybrids cause most of the confusion. A searchable scan is a raster image with an invisible OCR text layer placed over it. Its text can be selected and searched, but the characters came from recognition, not from the CAD system, and they can be wrong while the image underneath is right.

The opposite case is a vector PDF whose text was exported as geometry. Many CAD systems draw single-stroke drafting fonts as short line segments, so a dimension value arrives as strokes shaped like digits. The lines are exact, and there is no text layer to read. A third variant is a vector sheet with pasted images: a scanned title block, a raster detail view or a stamped revision box.

Text layers versus OCR

In a native text layer, “Ø12 ±0.05” is stored as characters. Reading it is a lookup: the value is exactly what the drafter's CAD system wrote, including the diameter symbol, the plus-minus sign and the decimal point. The value can still be wrong as engineering, because the drafter may have typed it wrong or overridden a dimension, but the act of reading adds no error.

OCR adds error at the characters engineering depends on most. These are the usual confusions on drawings:

• Decimal points lost or invented: 12.5 read as 125, or a speck of noise turning 125 into 12.5.
• The diameter symbol Ø read as 0 or O, so Ø8 becomes 08.
• ± read as +, which silently drops the lower half of a tolerance.
• 1, l and I; 0 and O; 5 and S; 8 and B, especially in part numbers and revision letters.
• Stacked tolerances and limit dimensions read as one line, or the upper and lower values swapped.
• Feature control frames read as loose text, with the compartment borders, symbols and datum letters lost or merged.

Drawing text makes some of this harder. Many drawings use single-stroke drafting lettering that general OCR engines were not trained on, and dimension text runs at angles along dimension lines. A value recognized from a scan is a candidate. Treat it as unproven until a person has compared it with the image. The guide to drawing interpretation errors covers the mistakes that survive this step.

Line weights and resolution

Drawing conventions use line width and line type to carry meaning: thick continuous lines for visible outlines, thin continuous lines for dimensions, extension lines and hatching, thin dashed lines for hidden edges and thin chain lines for centerlines. The ISO 128 series and ASME Y14.2 define these conventions. In a vector PDF the width is a stored number, although some exports write every line at the same hairline width and erase the distinction. In a raster file the width survives only as a count of pixels, so the scan resolution decides what is left.

The arithmetic is simple. At 300 dpi there are about 11.8 pixels per millimeter of paper. A 0.25 mm line is about 3 pixels wide, a 0.13 mm thin line about 1.5 pixels, and a decimal point, roughly as wide as the lettering's stroke, a few pixels. At 200 dpi the thin line is about one pixel wide, and scanner noise or compression can break it or erase it. A broken thin line is a lost dimension line, and a lost decimal point makes a value ten or more times too large.

What each one lets you extract reliably

From a native vector PDF

• Every text string exactly, with its position, size and rotation, so values, notes and title-block fields can be read without recognition error.
• Line and curve coordinates exactly, so views can be separated and shapes such as circles found in the geometry.
• Stroke widths and dash patterns, when the export kept them, to tell outlines from hidden, center and dimension lines.

What it does not give you is meaning: which text belongs to which feature, which view is which, what the hidden lines mean in 3D, and whether the drawing is complete. Those are interpretation, and they need explicit rules or a person who reads drawings. A value the drawing does not state is still missing in a perfect vector file; the guide to missing dimensions covers what to do then.

From a scanned or raster PDF

A person reads a good scan nearly as well as the original. Software can produce candidate text and candidate lines, and their quality falls with low resolution, skew, noise, lossy compression, handwriting, stamps and fold marks. Every value from a raster drawing that will drive geometry needs a human check against the image. A value the scan cannot show clearly needs the original sheet or a question to the drawing's owner.

How to tell which one you have

Preparing drawings for conversion

Most conversion problems are decided before the file is sent. The checklist below applies whether a person, a conversion program or a reconstruction service will read the drawing, and whether the target is a 2D file or a solid (the comparison of PDF conversion and engineering reconstruction explains the difference).

• Export from the CAD system directly to PDF. Avoid printing to paper and scanning, printing through an image driver, and flatten or optimize settings that rasterize the page.
• Keep text as text. Choose the export option that writes fonts rather than strokes or outlines, and keep symbols such as Ø and ± as characters.
• Keep line weights. Use a plot style or export setting that writes the drawing's line widths, so outlines, hidden lines and centerlines stay distinct.
• Put one part on one sheet where possible, with the title block filled in: units, material, general tolerance note, projection symbol, revision and drawing number.
• Send the current revision and check the file against the released drawing. A clean export of an out-of-date revision is a clean wrong drawing.
• Keep every dimension the part needs on the sheet. Values that live only in the CAD model or a separate document are invisible to anyone reading the PDF.
• If only paper exists, scan so that thin lines stay several pixels wide (300 dpi is a common minimum for line drawings, and small text benefits from more), straight, in grayscale or black and white, without lossy compression. Keep the original sheet for checking.
• Record where each value came from. A value read from a text layer, a value recognized by OCR and a value someone answered are three different kinds of evidence.