How to Read MTF Curves: Contrast, Spatial Frequency, and Sensor Matching for Machine Vision
MTF is contrast transfer versus spatial frequency. This guide covers both chart formats, sagittal and tangential curves, field-position data, sensor matching by pixel pitch, and the difference between design MTF and measured MTF.
An MTF curve plots how much contrast a lens transfers from scene to sensor, on a scale of 0 to 1, against spatial frequency in line pairs per millimeter (lp/mm).
To read one: calculate the Nyquist frequency of your sensor, read the lens contrast across the frequency band your task actually samples, which we start at 0.5× to 0.7× of that frequency, at the field positions your algorithm uses, and check that the sagittal and tangential curves stay close together. A lens that looks strong at the center can still fall short at the edge of the image circle, so field-position data matters as much as the on-axis curve.
What is MTF?
MTF, the modulation transfer function, is the ratio of image contrast to object contrast as a function of spatial frequency, measured in line pairs per millimeter (lp/mm). An MTF of 1.0 means the lens transfers full contrast at that frequency, while 0.3 means only 30% survives. Every MTF value is tied to a specific frequency, field position, and test condition.
One line pair is one dark bar plus one adjacent bright bar. A spatial frequency of 50 lp/mm packs 50 such pairs into one millimeter on the sensor, so higher lp/mm means finer detail. A 10µm line width in a periodic pattern, 20µm per pair, is 50 lp/mm; a 5µm line width is 100 lp/mm. An isolated feature has no single frequency and does not convert this way.
MTF is field-dependent: the same lens usually measures higher at the center than at the edge, so a full characterization reports curves at several image heights.
How do you read an MTF curve?
Identify the plot format first. MTF vs spatial frequency fixes the field position and sweeps detail fineness. MTF vs image height fixes the frequency and sweeps from center to corner. Read contrast at the frequencies your sensor samples and the field positions your algorithm uses, on both curves.
A datasheet with only one format is half a specification. Commonlands publishes both for its measured M12 lenses.
Reading the frequency axis
On an MTF vs spatial frequency chart, the x-axis runs in lp/mm and the y-axis runs from 0 to 1.0, or 0 to 100% on some datasheets. The information is in how fast the curve falls and where it sits at the frequencies you care about.
Datasheets usually overlay curves for several field positions: on-axis, around 0.7 field (70% of the lens's specified maximum image height), and full field. That maximum is the image-circle radius, not necessarily half the sensor diagonal. Each position carries sagittal and tangential traces. Higher and flatter is better across the band your application samples. The sensor-matching section derives that band from pixel pitch.
What do sagittal and tangential MTF curves mean?
Sagittal curves plot contrast for features oriented radially, along the line from the image center outward. Tangential curves plot contrast for features perpendicular to that radius. The two are identical on-axis for a perfectly centered, rotationally symmetric design.
A wheel makes it concrete: sagittal features run like spokes; tangential features run like the rim. Off-axis in an astigmatic lens, spoke-oriented and rim-oriented detail focus at slightly different distances, so one orientation is sharp while the other is soft at a single focus setting.
The separation is a diagnostic signal. On-axis, an S-T split points to assembly asymmetry such as decenter or tilt, though test-fixture misalignment can produce the same signature. Off-axis, a widening gap indicates astigmatism building toward the image boundary: round holes image as ellipses near the corners, and edge thresholds tuned at one orientation misfire at another. When feature orientation is fixed, verify the lower of the two curves at your operating frequency.
How do you read MTF vs field-position curves?
An MTF vs field curve plots contrast at a fixed spatial frequency against image height, the distance from the optical axis in millimeters. The center sits at 0mm, while the maximum is the image height the lens is specified to cover, the radius of its image circle. Only a matched format puts the sensor half-diagonal at that endpoint. A flat curve means uniform performance. A steep drop usually marks where correction runs out.
Convert your sensor geometry to image height before reading. A 1/1.7" sensor has a diagonal of about 9.4mm, so its corner sits near 4.7mm image height. Format names do not map exactly to millimeters, so take the diagonal from the sensor datasheet, not the format label. The sensor size and lens compatibility guide covers image circle coverage and format math.
Two shapes call for different responses. A smooth, moderate decline that still clears your contrast threshold at the corner is normal for a well-corrected design. A cliff at a specific radius usually marks a design boundary, and no focus adjustment fixes the soft corners a sensor gets by reaching past it.
Confirm the cause before you accept that reading: defocus, field curvature, vignetting, sensor or target tilt, fixture misalignment, and assembly decenter bend a field curve the same way, and a through-focus sweep separates them. Wide-angle designs face the hardest field demands, so a 4mm lens covering 98° needs field MTF verified at several image heights, not a center measurement extrapolated outward.
How do you match lens MTF to sensor pixel pitch?
Calculate the sensor's Nyquist frequency, 1000 divided by twice the pixel pitch in µm, then read the lens MTF at every field position your algorithm uses. Commonlands starts at 0.5× to 0.7× of Nyquist and at an MTF near 0.3. Both are heuristics, not standards, so validate them against your task: the band that matters follows from your feature size and contrast, and whether the lens limits the system depends on the whole chain of sampling, sensor, ISP, noise, and task tolerance.
That band sits below Nyquist because the sampled response there is set by more than the lens: pixel aperture, a color filter array and its demosaic, an optical low-pass filter, sampling phase, noise, and the ISP. A monochrome sensor with no OLPF holds contrast much closer to Nyquist, at the price of aliasing.
For a 1.85µm sensor, that band spans roughly 135 to 190 lp/mm. Against the 0.3 starting point, a lens that clears it there carries contrast the sensor can digitize, while one near 0.2 at half-Nyquist leaves those extra pixels resolving less detail than their count implies. Both readings are starting calls, not verdicts.
Setting the contrast threshold
No universal MTF threshold guarantees reliable machine vision, and an application name alone does not fix an image-plane frequency. That comes from the smallest feature the task reads, carried into the image plane by magnification, and it shifts with field position, spectrum, and scene contrast. Fix those first. The rows below are Commonlands starting heuristics, not published engineering standards; derive your own from feature size, magnification, field, wavelength, aperture, and a task acceptance criterion.
| Application | Typical image-space frequency (lp/mm at the sensor) | Starting MTF target | Notes |
|---|---|---|---|
| 1D barcode reading | 10–40 lp/mm | 0.4 at the narrow-bar frequency | Wide bars are forgiving; high-density Code 128 narrow bars are not |
| OCR | 40–80 lp/mm | 0.3 at the stroke frequency | Printed character strokes typically span roughly 100–400µm on the object, depending on point size and print process |
| Surface defect inspection | Set by minimum defect size | 0.3 at the defect frequency | Defect size × magnification gives feature size on sensor, which gives frequency |
| Dimensional measurement | 50–150 lp/mm | 0.4 at half Nyquist, uniform across the field | Sub-pixel edge detection also requires the distortion budget checked separately |
| AI / CNN object detection | 20–60 lp/mm | 0.3 at mid frequencies, field uniform | CNNs tolerate soft optics better than edge-based code, but train/deploy optics should match |
Relating image-space frequency to feature size in the scene
MTF is specified in the image plane, so object-space features must be converted through magnification. Image-space frequency equals object-space frequency divided by the magnification m (image size over object size). At m = 0.1, a 10 lp/mm pattern in the scene lands at 100 lp/mm on the sensor, so that is the frequency to read. The field of view calculator gives the magnification for your geometry.
Aperture belongs in the same calculation: stopping down one to two stops often lifts off-axis MTF as aberrations shrink faster than diffraction grows. On the fixed-aperture M12 lenses Commonlands measures, that operating point is set at purchase. A C-mount iris can be tuned at the fixture.
What is the difference between design MTF and measured MTF?
Design MTF is computed from the lens prescription in optical design software, with every element at its nominal position. Measured MTF is instrument data from a physical lens. It includes the decenter, tilt, spacing error, and glass variation a real build carries, so as-built MTF is typically lower than the design curve. Tolerance analysis predicts a distribution rather than one line. Two units of the same design measure differently, so acceptance should use measured data.
A standard report covers MTF vs spatial frequency and MTF vs field at 9 field points across 3 azimuths, plus EFL and distortion at 21 points. It also covers lateral and longitudinal chromatic aberration at 480nm, 546nm, and 644nm. Astigmatism and field curvature are included, with through-focus MTF on request.
Curves are measured on a Trioptics ImageMaster HR2, with calibration documentation available on request, so a report can back incoming inspection and lot acceptance. The $199 report is open to customer-supplied lenses, and near-infrared users can request 850nm and 940nm. Current scope and pricing are on the MTF testing service page.
A curve at F/2.0, infinity focus, 546nm cannot be compared directly against a curve at F/5.6, 500mm conjugate, broadband white light. Both are valid measurements of different operating points. A vendor showing only center MTF, wide open, at one wavelength is presenting the most favorable number, not the number your fixture will see.
MTF measurement options compared
Three routes produce MTF data, and they do not measure the same thing. Pick the cheapest route that isolates the variable you need to qualify.
| Approach | What it measures | Relative cost | Best fit |
|---|---|---|---|
| In-house slanted-edge (ISO 12233 chart plus analysis software) | System-level MTF: lens, sensor, and ISP together, at your working distance | Lowest; a printed target and software on hardware you already own | Pass/fail checks on your own production line |
| Commonlands MTF testing on the Trioptics ImageMaster HR2 | Component-level lens MTF on a calibrated bench, vs frequency and vs field at 9 positions across 3 azimuths | $199 per lens, no capital outlay | Incoming inspection or lot acceptance without owning a bench |
| Dedicated lab equipment or bench time from Trioptics or Optikos | Component-level MTF plus wider optical metrology, scaled to your throughput | Highest; capital equipment or per-hour bench service | High-volume incoming inspection or an internal metrology lab |
MTF-tested lenses and the MTF testing service
Commonlands measures lens designs on the Trioptics ImageMaster HR2, so selection decisions can rest on measured curves rather than design plots. Three MTF-tested options cover the demand space: the CIL560 4mm C-mount for wide-field coverage, the CIL561 6mm C-mount for a tighter field on the same 1/1.7" format, and the CIL122 IR-corrected 12mm M12 for near-IR work. The test service applies the same instrument to lenses you supply.
Commonlands stocks a broad range of M12 lens variants in the US, with C-mount and filter inventory alongside. Orders placed before 12 PM PT ship same day from San Diego, CA. ISO 9001:2015 certified.
Frequently asked questions
What is MTF in a lens?
MTF, the modulation transfer function, is the ratio of image contrast to object contrast at a given spatial frequency, expressed in line pairs per millimeter (lp/mm). An MTF of 1.0 means full contrast transfer, while 0.3 means only 30% survives. It is a curve across frequency and field position, not a single sharpness score.
How do you read an MTF chart?
Identify the format first: MTF vs spatial frequency plots contrast against lp/mm at fixed field positions, while MTF vs image height plots contrast against field position at fixed frequencies. Read the contrast at the frequency band your sensor samples, at the field positions your algorithm uses, and compare sagittal and tangential curves for astigmatism.
What does sagittal vs tangential mean on an MTF plot?
Sagittal curves describe contrast for features oriented radially, like spokes pointing at the image center. Tangential curves describe features perpendicular to the radius. They are identical on-axis for a centered design. On a measured unit, an on-axis split points to asymmetry in the assembly such as decenter or tilt, though test-fixture misalignment can produce the same signature. A large gap off-axis indicates astigmatism: round objects image as ellipses, and edge sharpness depends on orientation at that field height.
How do you match MTF to pixel pitch?
Calculate the Nyquist frequency: 1000 / (2 × pixel pitch in µm). A 1.85µm pixel gives 270 lp/mm. A 3.45µm pixel gives 145 lp/mm. Then check lens MTF across the band your task samples, at the field positions your algorithm uses. Commonlands starts near 0.5× to 0.7× Nyquist and at an MTF near 0.3; both are heuristics to validate against your task, not standards, and final resolution depends on the whole imaging chain.
What is the difference between design MTF and measured MTF?
Design MTF is the prediction from the optical prescription in design software, with every element perfectly made and positioned. Measured MTF comes from an instrument testing a physical lens and includes manufacturing tolerances such as decenter, tilt, and spacing error. As-built lenses typically measure below the design curve, so acceptance decisions should use measured data.
Need MTF data for a specific lens?
Send the Commonlands engineering team your sensor, pixel pitch, working distance, and the field positions your algorithm uses. We will pull measured curves or run a Trioptics HR2 report on the lens in question.