For calibrated inspection imaging, pixels per millimeter equals the measured pixel span divided by the same span in millimeters: 500 pixels across a 100 mm scale gives 5 px/mm.
Use the calibrated value in each image direction, distinguish sampling resolution from measurement accuracy, and verify it with a traceable reference before setting inspection limits.
Choose the correct pixels-per-millimeter calculation
The direct calibration formula is Pixels per millimeter = pixel count ÷ real distance in millimeters. A 100 mm reference spanning 500 pixels gives 5 px/mm, while its inverse, millimeters per pixel, is 100 ÷ 500 = 0.2 mm/px [S5].
The inverse is equally important when sizing a field of view: millimeters per pixel = millimeters ÷ pixels, and pixels per millimeter = 1 ÷ millimeters per pixel. At 0.5 mm/px, a 1 mm feature covers 2 pixels; at 0.2 mm/px, it covers 5 pixels [S5].
A pixel has no inherent physical size. Its scale depends on calibration or on an assumed pixel density, so a raw 1920 × 1080 image does not define millimetre coverage by itself [S2].
Derive the value from a physical reference
Place a calibration artifact in the same imaging plane as the inspected part, with its length traceable to an accepted dimensional standard. Measure the artifact between defined endpoints, count the corresponding pixels, and apply Pixels per millimeter = pixel count ÷ artifact length [S5].
For example, a 50 mm gauge spanning 250 pixels establishes 5 px/mm. A 25 mm gauge spanning 100 pixels also establishes 5 px/mm. These values support a consistent scale, while a 25 mm gauge spanning only 80 pixels yields 3.2 px/mm and signals a different or invalid setup.
Repeat the calculation horizontally and vertically. Square pixels commonly produce equal values, but unequal horizontal and vertical values are possible with non-square pixels or a non-orthogonal image transformation [S2].
Do not rely on screen pixels, monitor dimensions, or nominal sensor specifications as workpiece calibration. The converter labels its input as DPI, but uses it as pixels per inch; at 96 PPI, 100 pixels correspond to 26.4583 millimetres [S3].
Convert calibration into inspection coverage
After establishing 5 px/mm, the image width in millimetres equals image width in pixels ÷ 5. A 4096 px image then covers 819.2 mm, subject to lens distortion and edge effects that calibration must account for. [S5]
For defect sizing, the smallest measurable feature is represented by a number of pixels. At 5 px/mm, a 1 mm feature occupies 5 pixels and a 0.2 mm feature occupies 1 pixel. A one-pixel measurement is generally unsuitable for reliable acceptance because subpixel edges, noise, illumination variation, and segmentation uncertainty can dominate the result.
The theoretical reference uses a boundary of 2 pixels per line pair, and the calculator reports both that pitch and the pitch after applying the selected sampling margin [S4]. This is a planning relationship, not a guarantee of captured contrast.
Compare the main scale-setting methods
The Caltrans Surveys Manual gives photographic scale as focal length divided by height above terrain [S5].
Use the artifact method for a calibrated bench inspection. Use camera geometry only when object distance, sensor pitch, focal length, and orientation are controlled. Use PPI conversion for declared output density rather than metrology-grade dimensional calibration [S1][S3][S5].
Separate resolution, detection capability, and accuracy
Pixels per millimetre describes sampling density, not positional accuracy.
Flat-field illumination, focus, sensor noise, edge sharpness, and the imaging chain determine whether fine pixels contain useful defect information. Spatial-frequency conversion alone cannot predict modulation transfer or distinguish every defect at the Nyquist boundary [S4].
Control failure modes during inspection
Symptom: the scale differs between the reference and the measured part. Root causes include camera-to-part standoff variation, non-parallel surfaces, lens distortion, or an incorrect working distance. Corrective action is to calibrate at the actual working plane and geometry, then repeat at the field-of-view extremes. [S5]
Symptom: horizontal and vertical scales differ. Root causes include image rotation, anisotropic resampling, non-square pixels, or different magnification on each axis. Corrective action is to measure both directions and maintain separate scale factors when the error exceeds the inspection acceptance tolerance.
Symptom: nominal PPI produces the wrong physical size. Root cause is treating an image-density declaration as workpiece calibration. Corrective action is to replace the assumption with a measured reference and verify the resulting field of view before production use.
Symptom: a defect changes size after processing. Root causes include sharpening halos, denoising that removes weak edges, compression artifacts, or thresholding that alters boundary pixels. Corrective action is to validate the complete processing chain against known defect samples; do not repair the result by changing calibration scale alone.
Do not adjust pixels per millimeter to force a borderline measurement into acceptance. If the feature is not resolved by the established sampling plan, improve optical resolution or inspection geometry, then recalculate the coverage.
Apply the calibration in inspection planning
Set the required sampling density before selecting sensor area and optics. For a defect of minimum dimension d, the theoretical image span is d × pixels per millimetre. A 0.2 mm crack at 5 px/mm spans 1 pixel, while at 20 px/mm it spans 4 pixels, making edge extraction more defensible. [S4]
Use a traceable artifact and repeat measurements at multiple field locations to determine whether one global scale is acceptable. For airborne inspection, use an altitude and lens configuration compatible with the required GSD, and verify that relief, tilt, and calibration control do not dominate the result [S5].
A practical acceptance record should identify the artifact length, traced endpoints, pixel span, calculated px/mm, mm/px, camera configuration, working distance, lens, and verification result. Revalidate after changing magnification, focus, sensor crop, optical path, image orientation, or processing geometry.
For comparison with automated machine-vision inspection, the decisive inputs are the smallest defect, required field of view, calibrated px/mm, acceptable localization error, and processing resolution; this inspection-station cost guide addresses the surrounding hardware decision, while the calibration calculation remains the governing measurement.