Machine Vision DISCUSSION

Which machine vision lighting should I use for shiny metal parts: ring, dome, bar or backlight?

Started by adamoumoussa machine vision lightingdark field illuminationbacklightdome lightspecular reflection
5 replies 248 views 6 participants
Latest activity · 30 Sep 2026

Which machine vision lighting should I use for shiny metal parts: ring, dome, bar or backlight?

adamoumoussa Machine Vision Forum
#1

I am inspecting machined aluminium parts for two things: the outer dimensions and the presence of a laser-engraved mark on the top face. With a white ring light around the lens, the surface shows bright hot spots, and the engraving appears in some images and vanishes in others depending on how the part is rotated.

How do people decide between the different light types for reflective parts, and can software such as adaptive thresholding fix an unevenly lit image instead?

Community replies 5

Re: Which machine vision lighting should I use for shiny metal parts: ring, dome, bar or backlight?

#2

Think of a shiny surface as a mirror: the camera sees a reflection of whatever lies in the mirror direction. With a ring light close to the lens axis, flat areas facing the camera reflect the LEDs straight back and appear as hot spots, while anything tilted reflects the light away and goes dark. The engraving scatters light diffusely, so its contrast against the mirror-like background flips with the local surface angle and the machining marks. That is why it appears and disappears with rotation.

Choosing a light is choosing where the light comes from relative to that mirror geometry, so that the feature you want differs in brightness from its surroundings every time.

Re: Which machine vision lighting should I use for shiny metal parts: ring, dome, bar or backlight?

#3

For the outer dimensions, use a backlight. Place a diffuse light panel behind the part so the camera sees a dark silhouette on a bright background. The surface finish no longer matters at all, and the edges have the highest contrast you can get. For accurate measurement of thick or rounded parts, a collimated backlight, ideally with a telecentric lens, avoids light wrapping around the edges and making the part measure small.

Keep the backlight only slightly larger than the field of view, or mask it; a much larger panel sends stray light around the part edges and softens them.

Re: Which machine vision lighting should I use for shiny metal parts: ring, dome, bar or backlight?

#4

For the engraving there are two standard approaches. Dark field: a ring of LEDs at a low angle to the surface. The smooth face reflects this light away from the camera and looks dark, while the rough engraved area scatters some of it toward the lens and looks bright. Diffuse on-axis or dome lighting does the opposite: light arrives evenly from all directions, the smooth face looks uniformly bright and the engraving looks darker.

Which works better depends on the finish. Low-angle lighting also lights up every machining mark and scratch, so on a rough-milled face a dome or coaxial light is often cleaner. Try both with real parts before buying.

Re: Which machine vision lighting should I use for shiny metal parts: ring, dome, bar or backlight?

#5

Wavelength and ambient light are the next thing to control. A single-colour light, red for example, with a matching band-pass filter on the lens, plus a short strobed exposure, makes the image nearly independent of room lighting. Shield the station from windows and overhead lamps in any case; many unstable inspections trace back to ambient light that changes during the day.

Polarisers are often suggested for glare. A polariser on the lens alone does little on bare metal; it needs a second polariser on the light, crossed at 90 degrees to the first, and the pair costs at least three quarters of the light, so test it before designing around it.

Re: Which machine vision lighting should I use for shiny metal parts: ring, dome, bar or backlight?

#6

Software can compensate for smooth unevenness, but not for missing information. An adaptive threshold (cv2.adaptiveThreshold) compares each pixel with the mean of its neighbourhood, so a gradual brightness gradient across the part is handled; choose a block size several times larger than the stroke width of the engraving. Flat-field correction, dividing the image by a reference image of a uniform target under the same light, does the same job more cleanly.

Saturated hot spots are a different matter: where pixels are clipped at 255 the engraving detail is simply not in the image, and no algorithm restores it. Fix those with the lighting geometry, then use software for the remaining gradient.

TEP COMMUNITY