Machine Vision Lighting Selection Basics

What Lighting Selection Is Actually Deciding

Lighting selection is not a search for the brightest source. It is the choice of a geometry and a wavelength that make the feature of interest differ from its surroundings in the image, and it has to be made before the lens and the camera are fixed.

Almost every inspection failure that is later blamed on the camera is a lighting failure that the camera could not recover, because a feature that produces no contrast at the sensor cannot be reconstructed by processing. The decisions below are ordered by how much they change the result.

Wavelength Against Material Response

The first decision is spectral. The response of the part to a given wavelength is a material property, so the same source that makes one defect obvious can make another invisible.

The practical approach is to rank the feature wavelengths first: what does the defect absorb, reflect or emit differently from the surrounding material. Only after that does the choice of a specific source make sense. Selecting a source first and then searching for an application usually produces a system that works on the demonstration part and fails on the production batch.

Where Each Spectral Band Earns Its Place

White sources are the default because they preserve colour information and suit applications where the defect is chromatic, such as a coating mismatch or a wrong part. They are also the least selective, which is a limitation when the defect is a monochrome texture difference.

Ultraviolet sources are used where the feature fluoresces or where the surface layer responds differently from the bulk. In the POMEAS range the UV sources are specified at 365 nm and 395 nm, which covers the two common fluorescence excitation bands.

Near infrared sources are used where the surface needs to be penetrated rather than examined: the POMEAS infrared sources are specified at 850 nm and 940 nm. Silicon sensors remain responsive at those wavelengths, so no special camera is required, but the sensitivity is lower than in the visible band and exposure has to be planned for it.

The measured spectral spread across the POMEAS lighting range covers 265 to 940 um, which is the full span from ultraviolet through visible to near infrared.

Geometry Against Surface Shape

The second decision is geometric, and it follows from the shape of the feature rather than from its material. A raised feature and a recessed feature respond differently to the same illumination angle, and a specular surface responds differently from a diffuse one.

Direct front lighting at a shallow angle emphasises relief: edges cast small shadows that the camera reads as contrast. The same source at normal incidence flattens relief and makes scratches and dents harder to see.

Dark field illumination, where the source is arranged so that only scattered light reaches the lens, is the standard choice for scratches, pits and edge chips on a polished surface. The direct reflection misses the lens by design, so the field stays dark until a defect scatters light into it.

Backlighting is the strongest option when the feature is a silhouette. The POMEAS telecentric parallel backlights are specified for fields from 30 mm to 265 mm, and the parallel form matters because a diffuse backlight produces a soft edge whose position depends on the threshold chosen, which is not acceptable for dimensional measurement.

Diffusion and Why It Is Not a Softening Option

Diffusion is often described as a way to soften the image. That description is misleading, because what diffusion actually does is change the angular distribution of the illumination, and the image only softens as a consequence when the surface is rough.

On a curved or machined surface the issue is that a single source direction produces a bright band where the surface normal happens to bisect the illumination and viewing directions. The band moves as the part rotates, and an inspection that depends on intensity thresholds becomes position-dependent.

A dome source replaces the single direction with an approximation of a uniform hemisphere, so every surface element within the dome's coverage receives light from a similar set of angles. The band disappears and the intensity reflects the surface reflectance rather than the orientation. The cost is a large drop in intensity and therefore in exposure budget, and a loss of the directional information that made relief visible.

The choice between dome and direct source is therefore a choice about which property the inspection needs: orientation-independent reflectance, or orientation-visible relief.

The Order That Avoids Rework

The sequence that wastes least time is: define the feature, rank its wavelengths, choose the geometry, then check the exposure budget, and only then choose the source hardware.

Exposure budget is the step that is most often left to the end and most often forces a restart. A narrow-band source concentrates its output in a band the sensor samples at lower efficiency; a diffused source spreads the same output over a hemisphere. Both reduce the light that reaches the sensor, and both interact with the exposure time that the motion of the part allows.

Checking the budget before the hardware is chosen converts a late surprise into an early constraint.

Strobing and the Exposure Time Available

When the part moves, the exposure time is bounded by the distance the part travels during the exposure. Lighting selection therefore cannot be finished without knowing whether the source will be continuous or strobed.

A strobed source concentrates the output into a short pulse, which freezes motion at the cost of requiring synchronisation with the trigger and a controller that can deliver the pulse current. A continuous source is simpler but forces the exposure to be shortened, which in turn demands more intensity.

The measured intensity range across the POMEAS lighting range is not directly comparable between source types because the figures refer to different measurement conditions. What does transfer is the relative cost: a diffused source and a strobed source both buy image quality with light budget, and the budget is finite.

Common Selection Errors and Their Symptoms

The most common error is choosing the source before defining the feature. The symptom is an inspection that requires image processing to recover contrast that should have been present in the raw image.

The second is using a diffuse source where directional relief was needed. The symptom is a flat image in which scratches are visible to the operator but not to the algorithm.

The third is using a directional source on a curved part. The symptom is an inspection whose pass rate depends on the rotational orientation of the part, which appears as intermittent false rejects that cannot be reproduced off-line.

The fourth is neglecting the exposure budget until the mechanics are fixed. The symptom is a system that works at low speed and degrades exactly when the line rate is increased, which is often blamed on the camera rather than on the light available.

Frequently asked questions

What is the first step in machine vision lighting selection?

Define the feature to be detected and rank the wavelengths at which it differs from its surroundings. The source choice follows from that difference. Choosing the source first produces a system that works on the sample and fails on production.

Should I use white light or a coloured source?

White light preserves colour information and suits defects that differ chromatically. A coloured or narrow-band source increases contrast for a specific material difference but discards the rest of the spectrum. Use white when the defect type varies, and narrow-band when the defect is known and its spectral response is the discriminator.

When is a dome light better than a ring light?

A dome light is better when the part is curved, machined or otherwise orientation-sensitive, because it removes the bright band that moves as the part rotates. A ring light is better when directional relief is the feature, because the dome removes exactly that information. The cost of the dome is a large reduction in intensity.

Why does a diffuse backlight give a worse measurement than a telecentric backlight?

A diffuse backlight produces an edge whose apparent position depends on the intensity threshold used to find it, so the measurement changes with the threshold. A telecentric parallel backlight produces a parallel beam, so the edge position is defined by the geometry rather than by the threshold.

What does ultraviolet illumination solve?

It excites fluorescence or interacts with a surface layer differently from the bulk. In the POMEAS range the sources are specified at 365 nm and 395 nm, which covers the two common fluorescence excitation bands. It is not a general contrast improver.

What does near infrared illumination solve?

It penetrates surface layers that are opaque in the visible band and reaches structure below them. The POMEAS infrared sources are specified at 850 nm and 940 nm. Silicon sensors remain responsive but with reduced sensitivity, so exposure has to be planned for it.

Can lighting fix a resolution problem?

No. Lighting controls contrast, not resolution. If the feature is smaller than what the lens and sensor can resolve, more light will not make it resolvable; the magnification or the sensor has to change.