Near infrared lighting for printed packaging inspection

Printed graphics can dominate an inspection image even when they have nothing to do with the defect you need to detect. Near-infrared (NIR) illumination may reduce that distraction, but it can also erase a required date code, registration mark or other printed feature. For packaging inspection, the decision should come from representative samples rather than visible ink color. You will need to compare illumination geometries, measure useful and unwanted contrast separately, and verify the complete optical chain before selecting hardware.

Define what must disappear and what must remain

Start with an inspection requirement for each feature, not a general request to remove printing. A crease, seal boundary, contamination spot and printed code may need different imaging conditions.

Separate the image content into three groups:

• Required features: the defect, edge or marking used to make the inspection decision.
• Distracting features: decorative printing or background patterns that should contribute less variation.
• Reference features: marks or boundaries needed to locate the inspection region or check package orientation.

A successful wavelength reduces distracting contrast while preserving both required and reference features. Suppressing graphics is not useful if the positioning algorithm then loses its reference mark.

Do not predict NIR behavior from visible appearance. Different ink formulations can look alike under room lighting but behave differently outside the visible band. Black-looking marks are not guaranteed to remain dark, and colored graphics are not guaranteed to disappear.

Build a sample set containing acceptable packages, known defects, weak markings and normal production variations. Include different print batches, substrates and suppliers where those variations occur in production. Record which features must remain detectable on every sample.

Choose front lighting or backlighting by the inspection task

Front lighting measures light returned from the package. Backlighting measures light transmitted through it. These are different experiments, even when you use the same nominal wavelength.

Use bar lights when you need to vary incidence direction and compare shallow versus steeper illumination. Keep the camera position fixed while moving the light, so you can attribute changes to illumination rather than perspective.

Use backlights when the measurement depends on transmission or silhouette. Leave enough illuminated area around the package to accommodate positioning variation; nominal package width alone is not a sufficient sizing criterion.

For contents viewed through packaging, the related NIR fill-level inspection guide addresses the transmitted-light problem more directly. The broader surface-versus-silhouette lighting comparison helps when the correct geometry is still uncertain.

Match the light, camera, lens and filter as one chain

A camera image depends on the overlap between illumination output, package response, lens transmission, filter transmission and sensor response. A suitable light cannot compensate for a camera configuration that rejects its wavelength.

Check the camera's spectral-response information and optical-filter configuration before ordering illumination. Monochrome operation alone does not prove that the camera has useful NIR sensitivity. If you are considering an existing color camera, review the IR-cut filter configuration guide before interpreting a weak image.

Focus with the actual inspection illumination. A lens focused under visible light may need adjustment under NIR illumination. Check the center and corners, then verify that the required package-height variation remains acceptably sharp. Do not assume that refocusing fixes every limitation in lens transmission or image quality.

For a transmitted-light trial, the LS-BKL-0416 is an 850 nm Infrared backlight specified for constant-on operation only. Treat it as a candidate for that geometry, not as evidence that a particular package will transmit enough light.

The LS-LXR-0225 is an 850 nm band-pass filter. Nominal wavelength alignment makes it a candidate to investigate, but does not establish complete optical compatibility.

Verify the filter's transmission curve, blocking performance, clear aperture and mechanical fit against the intended camera and lens. Angular effects can also matter when a filter receives a wide cone of rays. Browse lens accessories and the accessories finder only after confirming those requirements.

A band-pass filter can reduce unwanted wavelengths, but ambient light inside its passband still reaches the sensor. The optical-filter selection guide explains why spectral matching is more useful than choosing a filter by its visible appearance.

Run a controlled sample trial before optimizing algorithms

Use the same sample positions, inspection regions and defect labels across each trial. Save original images and acquisition settings rather than relying on screenshots with automatic display adjustments.

1. Capture a visible-light baseline. Record which graphics, defects and mandatory marks are detectable. This establishes what the proposed NIR image must improve without losing essential information.
2. Fix camera processing. Disable automatic exposure and gain during comparisons. Use consistent gamma and image processing, preferably linear data when calculating intensity-based contrast.
3. Test NIR front lighting. Change incidence direction while keeping the camera and sample stationary. Record glare, shadows and contrast at each position.
4. Test transmission where practical. Move to backlighting only if the package construction and inspection geometry permit it. Check folds, seams, overlapping layers and printed regions separately.
5. Compare exposure requirements. A common exposure gives an initial response comparison. Then optimize each candidate within the production exposure limit, recording gain and illumination settings.
6. Add the optical filter. Compare filtered and unfiltered images with the inspection light operating, then repeat with that light off. This helps distinguish useful signal from ambient contribution.
7. Repeat across the sample set. Include position, orientation and package-height changes. A result from one flat, perfectly placed sample is not a production qualification.

Change one variable at a time. Otherwise, an apparent wavelength improvement may actually come from a different angle, longer exposure or stronger image processing.

Use the lens and lighting test plan to organize the trial record. For deployment checks, the lighting stability validation guide covers variation beyond the initial sample comparison.

Measure useful contrast and unwanted contrast separately

For two reasonably uniform image regions, a simple normalized contrast estimate is:

This estimate ranges from zero to one for nonnegative intensities with a nonzero denominator. It is useful for initial comparisons, but it is not a detection guarantee. Noise, texture, blur and region placement also affect inspection performance.

Consider these hypothetical mean intensities from linear 8-bit images. The background has been held at a similar level to make the comparison easy to follow.

For the NIR image, the arithmetic is:

• Defect contrast: |70 - 170| / (70 + 170) = 100 / 240 = 0.417.
• Decorative contrast: |160 - 170| / (160 + 170) = 10 / 330 = 0.030.
• Required marking contrast: |145 - 170| / (145 + 170) = 25 / 315 = 0.079.

The decoration is much less distracting, while the defect remains distinct. However, the required marking has also become weak. That may disqualify NIR as the only acquisition condition.

Measure each feature against its local background, not a convenient background elsewhere on the package. Record intensity spread as well as the mean. Repeat across packages and positions, then validate actual detection and false-reject performance on samples excluded from initial tuning.

Budget spatial sampling and exposure before selecting output

A promising stationary image may fail at production speed. Check spatial sampling and motion blur before deciding that a light is suitable.

Suppose the smallest relevant feature is 0.30 mm wide and you choose an initial sampling target of 6 pixels across it. This is a trial target, not a universal inspection requirement.

The dimensional-inspection resolution guide explains why pixel count alone does not establish measurement capability. Use the calculator to check a different feature size or field of view.

Now assume the package moves at 500 mm/s and the initial blur allowance is 0.05 mm:

This is an estimate for constant motion during exposure. Trigger variation and mechanical movement can add other errors.

Evaluate the NIR image at that exposure, not only at a convenient bench setting. If contrast becomes unreliable, investigate illumination geometry, available optical throughput, sensor response and the exposure budget. Increasing gain may brighten the displayed image without restoring missing signal quality.

If pulsed operation is needed, select a light and controller documented for that operating mode. Confirm pulse duration, current, duty cycle and synchronization limits rather than treating a continuous-only light as a strobe.

Recognize when NIR is the wrong single-image answer

NIR is unsuitable as the only acquisition condition when required information depends on visible color differences. It can also fail when the defect and background respond similarly, even though distracting graphics disappear.

Use separate visible and NIR acquisitions when they provide complementary information and the cycle time permits it. A visible image might verify printing while an NIR image evaluates another feature. Each acquisition needs its own validated settings and decision criteria.

Sequential imaging introduces a registration problem on moving packages. Account for motion between exposures instead of assuming that identical pixel coordinates describe the same physical feature.

Do not assume that changing wavelength removes specular glare. Reflection geometry still matters. Polarization is also wavelength-dependent: use polarizers characterized for the intended NIR band rather than assuming a visible-light accessory will work.

If unwanted brightness persists after spectral filtering, investigate reflections and optical flare with the stray-light diagnostic guide. More filtering is not always the appropriate fix.

Check the evidence before ordering

Use this checklist to turn a promising demonstration into a defensible selection:

• Feature coverage: every mandatory defect, code and reference mark has a documented acceptance test.
• Sample coverage: print batches, material changes, folds and normal positioning variation are represented.
• Geometry: front lighting and transmission have been compared where both are physically practical.
• Spectral compatibility: camera response, light output, lens transmission and filter transmission overlap usefully.
• Optical performance: focus and feature visibility remain acceptable across the field and package-height range.
• Timing: exposure, trigger behavior and any light pulses meet the production motion budget.
• Operating mode: the selected light and controller support the intended continuous or pulsed operation.
• Verification: original images, settings and inspection outcomes are retained for independent validation.

Compare candidates in the lighting finder only after these requirements are clear. A sample image, field of view, minimum feature size, line speed and required marking list make a selection discussion much more productive.

Use the lighting configurator to assemble a candidate setup, or request application support with the trial record. The next decision should be based on preserved inspection information, not simply on how much printing disappears.