Machine Vision Illumination Guide

NIR Imaging for Machine Vision: 850nm vs 940nm, Bandpass Filters, and IR-Corrected Lenses

Why near-infrared reveals contrast that visible light misses, how to pick between 850nm and 940nm, and what the complete filter-lens-illumination stack requires.

By Max Henkart, Commonlands · Updated June 2026 · 10 min read

A board-level M12 camera ringed by near-infrared LEDs glowing dull red at dusk

850nm gives higher sensor quantum efficiency and a stronger signal for the same illuminator power, but the LEDs produce a faint visible red glow. 940nm produces no perceptible glow under normal conditions but needs more power or exposure time, because silicon sensor QE at 940nm runs at roughly half or less of its 850nm value.

A complete NIR system also needs a bandpass filter matched to the illumination wavelength and, if it must also image in visible light from the same focus position, an IR-corrected lens.

850nm vs 940nm: How to Choose the Right NIR Wavelength

Choose 850nm when a faint glow is acceptable and detection range or exposure time is the binding constraint. Choose 940nm when the illuminator must stay invisible, or when the system runs in bright daylight where suppressing ambient NIR matters more than sensor quantum efficiency. Commonlands stocks matched bandpass filters for both wavelengths.

850nm: higher sensor QE with a faint visible glow

Silicon CMOS sensors have higher quantum efficiency at 850nm than at 940nm, so for the same drive power an 850nm source gives more usable signal, supporting shorter exposures or lower illumination power. The tradeoff is glow: 850nm sits close to the edge of human vision, where deep-red sensitivity falls off steeply but does not vanish until roughly 750nm, and the short-wavelength tail of the LED's emission puts a small fraction of its output near that edge. At high drive currents this glow is detectable in dark environments.

940nm: no visible glow but roughly half the silicon QE

940nm sits far enough outside the eye's response that it produces no perceptible glow under normal conditions. The cost is sensor response: silicon QE at 940nm is commonly around half or less of its 850nm value, varying by sensor. A 940nm system therefore needs more power, longer exposure, or both to reach the signal-to-noise ratio an 850nm system achieves at the same range. High-throughput lines where exposure time is the binding constraint should verify this margin against the actual sensor QE curve.

Atmospheric and outdoor considerations

Outdoors, ambient sunlight carries strong NIR that competes with active illumination. Atmospheric water vapor absorbs solar radiation in a band around roughly 930nm to 970nm, which suppresses ground-level solar irradiance at 940nm relative to 850nm. That is a large part of why 940nm suits outdoor systems such as driver monitoring and face recognition: the ambient background is lower, partly offsetting the QE penalty. 850nm sees the full solar NIR floor.

Criterion 850nm 940nm
Sensor quantum efficiency Higher, with a stronger signal per watt of illumination Lower, commonly around half of the 850nm value or less; verify per sensor
Visible glow Faint red glow visible in dark environments No perceptible glow under normal conditions
Illumination power needed Lower for a given signal level Higher, to close the sensor quantum efficiency gap
Typical use case Industrial inspection, barcode reading, traffic monitoring where glow is acceptable Covert or human-facing installations where illuminator visibility is unacceptable
Matching single-bandpass filter CBP850 CBP940
Matching dual-bandpass filter CDB850 CDB941

These filters are not interchangeable: a 940nm bandpass filter blocks 850nm illumination and produces a dark image if paired wrong. See bandpass filter machine vision for how center wavelength and passband width factor into the selection.

A Commonlands 850nm bandpass filter held in tweezers, its coating showing a dichroic color sheen
The Commonlands 850nm bandpass filter passes only the 850nm band for clean NIR imaging.

What NIR Imaging Means in Machine Vision

NIR imaging uses wavelengths from roughly 700nm to 1000nm, just beyond visible red. In machine vision the term almost always means active illumination at 850nm or 940nm paired with a camera set to detect that band, not ambient infrared or thermal imaging.

Switching to NIR moves four elements of the optical stack together:

  • The IR-cut filter must be out of the optical path. It blocks wavelengths above roughly 650nm and stops NIR reaching the sensor.
  • The lens must transmit the NIR wavelength, and its focus plane shifts between visible and NIR unless the lens is IR-corrected.
  • A bandpass filter matched to the wavelength rejects ambient visible light and improves signal-to-noise ratio.
  • The sensor needs meaningful quantum efficiency at the target wavelength. Silicon QE drops significantly above 850nm.

Removing the IR-cut filter and switching on an LED, with nothing else changed, is not enough. The image comes out dim, soft, or noise-limited. Most machine vision NIR is active: a dedicated illuminator with a matched bandpass filter controls wavelength and intensity independent of ambient light. Passive NIR relies on ambient NIR already in the scene, so it works outdoors in daylight but fails at night and under modern white LED or fluorescent fixtures.

See the Commonlands image sensor selection guide for how NIR response weighs against resolution, pixel size, and shutter type.

Why NIR Reveals Contrast That Visible Light Misses

NIR imaging is useful because reflectance, transmission, and absorption vary with wavelength, so many materials look different in NIR than under white light. That difference is the entire basis for NIR machine vision, not a general-purpose upgrade to image quality.

Inks and printed graphics

Carbon-based inks absorb visible light and stay dark, and that absorption carries into the NIR band, so carbon-ink printing keeps its contrast against the substrate under 850nm or 940nm. Many dye-based inks that look opaque black to the eye are largely transparent in NIR, so a barcode printed in dye-based ink can effectively disappear in a 940nm image, revealing the substrate beneath.

That cuts both ways. Disappearing ink helps when inspecting a feature under a label, but it defeats an OCR or print-verification system reading dye-based characters. Carbon-based ink does not have this problem. Verify ink composition on the actual production substrate before committing to a wavelength.

Surface glare and coatings

Specular reflections from glossy or metallic surfaces saturate visible pixels when ambient light is not controlled. Narrow-band NIR with a matching bandpass filter rejects the broadband visible glare, so the sensor mainly captures the NIR scattered from the surface. It does not remove specular reflection from the NIR source itself, but it removes the visible glare that dominates typical factory lighting.

Thin films and coatings that look uniform under white light can show different NIR reflectance by composition or thickness. Biological and food-grade materials often show NIR contrast tied to water content, which is why NIR appears in fill-level inspection, food sorting, and pharmaceutical packaging.

What NIR does not do

NIR is not universally better than visible. Color discrimination and legibility of visible ink are usually better served by visible light. NIR does not penetrate opaque materials the way X-ray does, and useful penetration depth varies by material and must be validated on real samples. A datasheet contrast difference does not guarantee the same result on production parts.

Commonlands 850nm bandpass filter mounted in front of an M12 lens for NIR machine vision inspection
A Commonlands 850nm bandpass filter mounted ahead of the lens element rejects ambient visible light before it reaches the sensor.

Where NIR Imaging Shows Up in Production Systems

NIR imaging shows up wherever visible-light contrast fails on a specific material pairing, which clusters around a few recurring problems.

Traffic and license-plate systems are among the most common deployments. Retroreflective plate coatings return a strong signal under 850nm synchronized with a short exposure, giving high-contrast plates regardless of ambient light or headlight glare. Because these run day and night, they are the most common use for the Commonlands CLA216-ICR-850BP switcher: color video by day, 850nm NIR at night, one camera. See lenses for traffic monitoring for lens selection.

Barcode reading uses NIR when the code is carbon-based ink on an NIR-reflective substrate, or when ambient lighting is uncontrolled. Robotics platforms near people sometimes pick 940nm so the illumination stays invisible. Quality-inspection lines for packaging and seals use NIR to reveal defects invisible in the visible band, running single-bandpass under controlled illumination.

Filters, Lenses, and Switching Architectures for NIR Systems

Beyond sensor and illumination choice, an NIR system needs three hardware decisions: filter type, lens IR correction, and whether it must switch between visible and NIR modes at all.

Filter selection

The IR-cut filter is the most misunderstood element: it blocks NIR and must be out of the optical path, but removing it is necessary, not sufficient, because without a bandpass filter the sensor still sees ambient visible light. A single bandpass filter passes only the illumination band. A dual-bandpass filter passes a visible and an NIR window at once for RGBIR (red, green, blue, infrared) sensors but cannot fully reject either. An electronic switcher moves an IR-cut and a bandpass filter in and out for true day/night separation.

When lens IR correction matters

If a system runs only in NIR, focus at the NIR wavelength and IR correction is unnecessary. If it must stay sharp at both visible and NIR from one focus position (day/night cameras, RGBIR sensors, dual-mode inspection), an IR-corrected lens is required. A standard lens has chromatic focus shift between visible and NIR from dispersion in the glass. Removing the IR-cut filter does not close that gap, since the shift is in the lens design, not the filter.

Sensor selection

Not every sensor is a good NIR candidate. Many consumer and some machine vision sensors have an NIR-blocking layer in the pixel stack itself, independent of any external IR-cut filter, capping sensitivity at 850nm and 940nm. Pull the sensor's spectral QE curve and read the value at each wavelength rather than assuming a generic silicon response. RGBIR sensors depend on a dual-bandpass filter to define the NIR passband, or leakage degrades daylight color.

Commonlands NIR Imaging Components

IR-corrected M12 lenses, 850nm and 940nm single-bandpass filters, and an electronic filter switcher for NIR systems. The six Commonlands parts below cover 850nm and 940nm builds, ranked by the order in which each choice locks down the rest of the design.

Rank Component Type Key spec Best for
1 CIL122 IR-corrected M12 lens 12mm EFL, F/2.0, 1/1.7in 8-12MP Day/night and dual-mode systems that must stay sharp in visible and NIR from one focus position
2 CBP850 850nm single-bandpass filter Passes 850nm, blocks visible; 7.0mm circular, 0.3mm thick 850nm active illumination where a faint LED glow is acceptable and higher sensor QE matters
3 CBP940 940nm single-bandpass filter T>90% at 940nm, blocks the visible band Covert or human-facing 940nm systems where no visible glow is allowed
4 CLA216-ICR-850BP Electronic IR-cut / 850nm bandpass switcher IR-cut and 850nm bandpass in one moving holder, 7.6mm height Single-camera day/night with full rejection in each mode; pairs with the IR-corrected lens
5 CDB850 Dual-bandpass filter (visible + 850nm) Passes a visible window and an 850nm window in one fixed element Switcherless day/night on RGBIR sensors at 850nm, no moving parts
6 CDB941 Dual-bandpass filter (visible + 940nm) Passes a visible window and a 940nm window in one fixed element Switcherless day/night on RGBIR sensors at 940nm, no visible glow

The ranking follows build order, not price: chromatic focus shift originates in the lens glass and cannot be filtered out afterward, and a bandpass center-wavelength mismatch costs the most signal of any single choice. MidOpt (midopt.com) and Edmund Optics also sell bandpass and dual-bandpass lines outside the Commonlands filter range. Confirm the lens field of view covers the illuminator beam angle with the field of view calculator before ordering.

IR Cut Filter Edmund Optics

Reflective 650nm IR Cut Off Filter

$9.00

View Product
Blue Glass IR Cut Off Filter

650nm Blue Glass Reflective IR Cut Off Filter

$11.00

View Product
Midwest Optical BP850 850nm Bandpass filter CBP850

850nm Bandpass Filter

$9.00

View Product
940nm Bandpass Filter BN940 Midopt

940nm Bandpass Filter T>90%

$9.00

View Product

Browse Optical Filters for Cameras

A square Commonlands NIR bandpass filter mounted in front of an M12 lens on a camera board
Pairing a Commonlands bandpass filter with NIR light cuts ambient interference.

Frequently Asked Questions

What is the difference between 850nm and 940nm in machine vision?

850nm and 940nm are the two standard NIR illumination wavelengths. Silicon sensors have higher quantum efficiency at 850nm, so it gives a stronger signal for the same illuminator power, but the LEDs emit a faint visible red glow. 940nm shows no perceptible glow but needs roughly double the power or exposure to match, since silicon QE at 940nm is often half its 850nm value or less.

What is NIR imaging in machine vision?

NIR imaging uses near-infrared wavelengths, typically 850nm or 940nm, to illuminate and image a scene outside the visible band. A complete system needs a sensor with no IR-cut filter in the path, a bandpass filter matched to the wavelength, an NIR light source, and an IR-corrected lens if it must also image sharply in visible light from the same focus position.

When should engineers use NIR imaging instead of visible light?

Use NIR when visible light does not give enough contrast on the target: printed graphics where ink and substrate share similar visible reflectance, specular glare from glossy surfaces, coatings that look uniform under white light but vary in NIR, or unpredictable ambient lighting. NIR is not always better. Validate that the real materials produce the expected NIR contrast first.

What filters are used for NIR imaging?

Three filter types appear in NIR systems. A single bandpass filter at 850nm or 940nm passes only the illumination wavelength and blocks visible light. A dual-bandpass filter passes a visible window and an NIR window at once in one fixed element, used for switcherless day/night on RGBIR sensors. An electronic switcher moves an IR-cut filter and a bandpass filter in and out for true day/night separation.

Do I need an IR-corrected lens for NIR imaging?

If the system operates only in NIR, focus at the NIR wavelength and correction is not required. If it must also image sharply in visible light from the same mechanical focus position, an IR-corrected lens is required. A standard lens has chromatic focus shift between visible and NIR. Removing the IR-cut filter does not fix it, because the shift is in the glass, not the filter.

Need Help Designing an NIR Imaging System?

Describe the sensor, working distance, illumination wavelength, and inspection goal. Commonlands engineering can help you choose between 850nm and 940nm, match a bandpass or dual-bandpass filter, and pair it with an IR-corrected M12 lens.