Test targets for machine vision: USAF 1951, Ronchi, star and DOF

A test target turns an argument about image quality into a number. Benchmarking a new lens, troubleshooting a blurry station, or matching several cameras to one standard all come down to reading line pairs per millimeter (lp/mm) off a physical chart and deciding pass or fail. Four target families cover machine vision work: the USAF 1951 chart for discrete resolution numbers, the Ronchi ruling for best focus across the whole field, the star target for continuous resolution in every orientation, and the 45° telecentricity and depth-of-field targets for magnification error and focus hold. The formulas, the conversion arithmetic and one worked pass/fail example follow, so you can specify the target that answers your actual question.

Pick the target by the question you are asking

Buy the target that answers your question, not the one that looks most scientific. The decision in one line each:

• You must quote a resolution number at specific field points: USAF 1951 chart.
• You must find best focus across the entire field at your working frequency: Ronchi ruling.
• You must compare two lenses, or read resolution in every orientation at once: star target.
• You must know how far a part can sit from focus before detail dies: depth-of-field target at 45°.
• You must know how much magnification moves with depth on a gauging lens: telecentricity target at 45°.

Many stations need two: a USAF chart for the resolution figure and the 45° pair for the depth behaviour of a measurement lens. Physical targets, calibration boards and holders sit in the lens accessories category; the glass you will benchmark comes from lenses and telecentric lenses.

Reading the USAF 1951 chart

The chart packs bar patterns into groups of six elements. Each element holds a set of vertical bars and a set of horizontal bars; one line pair (lp) is one black bar plus one white space, and resolution is quoted in lp/mm at the target. Groups run from -2 to 7, and the group number G with the element number E fixes the frequency:

Group 1, element 3 gives 2^(1 + 2/6) = 2.52 lp/mm, matching the published tables, and the top of a standard chart, group 7 element 6, gives 2^(7 + 5/6) = 228 lp/mm. Full tables for every group and element are tabulated in the 1951 USAF resolution chart reference. Vertical bars measure horizontal resolution; horizontal bars measure vertical.

Two reading rules keep the number honest. The reading is the last element whose bars stay separable, judged by a contrast measurement in software, because the eye interpolates repetitive patterns. Set one contrast threshold and hold it for every position; the measurement ideas behind that threshold are covered in lens contrast and resolution testing. And evaluate every field position at a single focus setting: refocusing between center and corner tests two systems, while a deployed lens gets one focus.

A lens focused dead-center can be sharp in the middle and soft at the corners. A slight defocus that balances the field costs center sharpness and is usually the right trade when the application's detail sits off-axis.

Ronchi rulings: one frequency, the whole field

A Ronchi ruling is a plate of parallel lines at a single spatial frequency covering its entire surface. That one property makes it the best-focus instrument: with the pattern filling the field, center and all four corners sharpen in the same image, and the position where the whole field is equally crisp is your balanced focus. Astigmatism blurs one orientation only, so a single orientation hides it; rotate the ruling 90° and retake the image to expose it. Because the ruling is a fixed frequency, tilting it to 45° also turns it into a depth-of-field target on the spot.

The cost is coverage. A 10 lp/mm ruling answers questions about 10 lp/mm; a different frequency needs a different ruling. Use the ruling for focus and the USAF chart for the resolution figure.

Star targets: continuous resolution in every orientation

Each element of a star target is a circle of alternating wedges tapering toward the center at a known angle. Walking from the rim inward is a continuous frequency sweep, and the wedges run in every orientation at once, so astigmatism and orientation-dependent blur show themselves without rotating anything. Stars scattered across the plate let you judge center, mid-field and corner in one image, which makes it the fairest lens-to-lens tool: shoot both lenses at the same working distance, f-number and lighting, and the corner stars settle the argument.

The price is readout. Frequency varies continuously with radius, so the exact lp/mm where the wedges merge takes measurement software, and simple line profiles do not follow the circular geometry. Stars rank systems superbly; they quantify only with real analysis.

Telecentricity targets: keystoning in degrees

A telecentricity target is a plate of vertical lines viewed at 45° to the optical axis, bottom edge farther from the lens. A non-telecentric lens images the far lines closer together, so a line's center at the bottom of the image sits displaced from its position at the top. A telecentric lens blurs the lines symmetrically instead, and the centers hold still.

Measure it with two horizontal line profiles:

where dY is the pixel distance between the two profiles and dX the shift of one line's center between them. Example profiles 400 px apart with a 3.5 px shift give atan(3.5/400) = 0.5°.

That angle has a direct cost: an edge on a part Z millimetres off focus is displaced Z × tan θ in the image. At 0.5°, a part 3 mm off focus shows its edges 26 µm out of position (3,000 µm × tan 0.5° = 26 µm). Set the budget from your presentation: with parts arriving anywhere in a ±3 mm band and a 10 µm measurement budget, the pass mark is atan(10/3,000) = 0.19°, and 0.5° fails it outright. For gauging at unit magnification, the LS-TCL-1972 (1x, WD 440 mm, 2/3", C-mount) is the class of lens this test validates.

The angle you measure here is the same quantity a data sheet calls telecentricity; telecentric lens distortion and telecentricity covers how the two relate.

Depth-of-field targets: the 45° tilt is a depth converter

A depth-of-field target is a ruling or frequency ladder tipped at a known angle, 45° in the standard set-up, so one image sweeps focus through depth. Where the bars stop being resolved along the tilted rule marks the depth limits: measure the in-focus run's length on the target and multiply by the sine of the tilt angle. At 45° that factor is 0.707, so a 10 mm in-focus run is 7.1 mm of usable depth. A steeper tip packs more depth into the same plate; 45° is simply the convention.

Work at the frequency your application needs and judge the limit with the same software contrast threshold you used for resolution. The result is a DOF figure for one f-number and one focus setting, which is exactly the trade against brightness described in working f-number, light and depth of field; on gauging lenses, validate it as shown in telecentric depth of field.

Worked example: passing a 0.3x lens with a 5 MP camera

The case: a station images small parts backlit, with a 5 MP monochrome camera with 3.45 µm pixels. The question: does the lens deliver what the camera can sample, and what is the pass mark on the chart?

Inputs:

• Camera: 3.45 µm pixels (example input; substitute yours).
• Lens under test: the LS-FFL-0914, a 2/3" C-mount machine vision lens, 0.3x, WD 168 mm.
• Target: USAF 1951 chart, read with a fixed software contrast threshold.

The rule: a sensor samples at most one line pair per two pixels (the Nyquist limit), and object-space frequency maps to sensor frequency divided by magnification.

Arithmetic:

1. Sampling ceiling at the sensor: 1 / (2 × 0.00345 mm) = 145 lp/mm.
2. Same ceiling in object space at 0.3x: 145 × 0.3 = 43.5 lp/mm. On the chart, group 5 element 3 (2^(5 + 2/6) = 40.3 lp/mm) is the last element safely below it; group 5 element 4 (2^(5 + 3/6) = 45.3 lp/mm) is already past it.
3. The reading: the software's last resolved element is group 3 element 4, which is 2^(3 + 3/6) = 11.3 lp/mm on the target.
4. Convert to the sensor: 11.3 / 0.3 = 37.7 lp/mm, so one line pair spans 1,000 / 37.7 = 26.5 µm, which is 7.7 pixels of 3.45 µm per line pair.

Decision: 37.7 lp/mm is 26% of the 145 lp/mm ceiling, so sampling is nowhere near the limit; the lens and focus setting are what you are measuring. The pass mark for this station: if the finest feature you must see is an 88 µm line pair (1,000 / 11.3 µm), this system resolves it cleanly and passes. If the job needs 23 µm line pairs on the object, stop lens shopping: 43.5 lp/mm is the camera's object-space ceiling at 0.3x, so you need more magnification or smaller pixels, and the object space pixel size arithmetic reaches the same verdict from the other direction.

How to choose and what to check before ordering

Match the purchase to the job:

• Quote a number: USAF 1951 chart; take the multi-position variant if corners matter.
• Balance focus at one frequency: Ronchi ruling matched to your detail. A 0.1 mm line-pair feature is a 10 lp/mm ruling, so pick accordingly.
• Compare lenses: one star plate, both lenses, same light and working distance.
• Validate a gauging lens: the 45° telecentricity and DOF pair.

Check two things before ordering. The pattern area must exceed your field of view at the working distance; the margins question is the same one covered in calibration target field of view and coverage margins. And the chart's frequency range must bracket your sampling ceiling with room to spare. Calibration boards and depth-of-field rulers ship as one family; the LS-LXR-0029 is a dot grid with a 0.5 mm dot diameter on a 76 × 76 mm board, the scale side of the same bench work.

When the usual target is the wrong tool

• Reading by eye and quoting the result: the number belongs to the viewer, not the system.
• A USAF chart left centered, when the application's detail lives in the corners.
• A single-frequency Ronchi ruling bought as the resolution test: it tests one frequency.
• A star target without analysis software: it ranks, it does not read out.
• Fine groups that resolve beautifully on a low-pixel camera: bars past the 1/(2 × pixel size) ceiling are aliasing, and aliasing looks like resolution. Compute the ceiling first.
• Refocusing between field positions: one focus setting, or the test no longer reflects the deployed system.

A short checklist before you benchmark

• Compute the sensor ceiling 1/(2 × pixel size), then the object-space ceiling at your magnification, before touching a target.
• Fix one contrast threshold in software and use it for every field position.
• One focus setting for the whole test; defocus slightly to balance the field if corners matter.
• USAF chart: slide to center, edge and corner; record group and element at each position.
• Ronchi: rotate 90° if one orientation looks softer.
• 45° targets: convert the in-focus run × 0.707 to depth; convert dX/dY to degrees with atan.
• Log f-number, working distance and lighting with every image.

When the benchmark points at the glass, start from the lens finder or the configurator's lens step. If you want a second opinion on a target plan for a specific camera, use the contact page; free expert support comes with every order.