Ghost Edges on Glass and Transparent Parts: Causes and How to Intercept Them

A glass part does not have one edge per feature. It has at least three: the entry face where light goes in, the exit face where it comes out, and the wall of any hole or bore between them. All three image, at different depths and with different contrast. What an operator calls a "ghost edge" is usually not a ghost at all — it is a real image of a real surface the algorithm was never told to ignore.

That distinction matters because the remedies differ. A true ghost from a reflection between two surfaces is removed by changing the geometry or adding a wedge. A real image of the back face of a cover glass is not removed by anything optical; it is removed by choosing which surface the measurement refers to and making the others invisible enough that only one edge survives thresholding.

The second thing worth establishing early is that glass edges are low-contrast in reflection and high-contrast in transmission. Most stations that struggle with ghost edges are using the wrong illumination geometry.

Why transparency produces more edges, not fewer

An opaque part presents one silhouette. A transparent part presents the silhouette of every interface where the refractive index changes: front surface, back surface, bore walls, chamfers and any internal feature, each at a different depth.

With a finite numerical aperture the lens accepts a cone of rays rather than a single ray, so the front and back surfaces of a thick part image at slightly different lateral positions as well as depths. Telecentricity removes the depth-dependent magnification change, which is a large part of the problem on an entocentric lens, but it does not make the second surface disappear.

The six interception points for ghost and virtual edges

1. Decide which surface the dimension refers to

This sounds like a documentation question and it is the most common root cause. If the drawing dimension is the hole diameter at the entry face and the station measures the exit face, the two differ whenever the bore is tapered or the part is not parallel to the sensor.

Write down which face the measurement refers to, then confirm it by imaging the part from both sides. On a through-hole in glass the entry and exit diameters are frequently not the same number, and the difference is real.

2. Compare the part thickness against the published depth of field

The front and back faces are separated by the part thickness. If that exceeds the depth of field, one face is defocused, and a defocused edge is a wide, low-gradient feature a threshold will locate unreliably — which is what a drifting or doubled edge looks like.

Published POMEAS depth of field figures relevant here: PMS-LTCM1-110 at 0.89 mm, PMS-LTC2-110C at 0.5 mm, PMS-LTC1-110C at 1.6 mm at 1:1, and PMS-LTCM023-90 at 7.25 mm. Where the glass is thicker, either a lower-magnification model is needed or the measurement has to be taken from the focused face with the other suppressed by illumination.

3. Use transmission, not reflection

A transparent part imaged in reflected light produces weak edges from both surfaces and strong specular highlights from any chamfer. In transmitted light it produces a silhouette, and the bore wall is the only thing competing with the true edge.

The POMEAS VP-GLTD telecentric parallel LED backlight series is built for this: beam diameters of 30 mm, 60 mm, 100 mm, 145 mm, 189 mm and 265 mm, working distances covering 45–500 mm, and red, green, blue and white options. Wavelength matters on glass because transmission and edge contrast are wavelength-dependent.

4. Set the illumination numerical aperture deliberately

This is the most effective interception point, and it is a lighting decision rather than a lens decision. A diffuse backlight sends light at every angle, so it wraps around the edge and produces a gradual intensity ramp instead of a step. A collimated backlight sends light in one direction, so the transition is sharp.

The VP-GLTD product page states this directly: compared with an ordinary diffuse backlight, parallel illumination reduces the boundary blur caused by light diffracting or spreading around the workpiece edge, raises edge contrast, and helps the telecentric lens exploit its depth of field and telecentricity.

The practical check is to compare the intensity profile across the edge under a diffuse source and under a collimated one; on glass the difference is usually obvious in a single greyscale plot.

5. Watch the bore wall acting as a light pipe

In a through-hole in glass the cylindrical wall between the faces can guide light and appear as a bright or dark ring inside the hole. The apparent diameter then depends on the wall geometry, part thickness and illumination angle rather than the actual bore.

Collimated light reduces this, since less light enters the wall at grazing angles. Where it does not, the answer is to measure at a defined depth, which means the depth of field has to be small enough to isolate that plane, pulling against check 2.

6. Check the coaxial path if the lens has one

Coaxial telecentric lenses such as PMS-LTC2-110C and PMS-LTC1-110C carry a built-in beam-splitter so illumination can be delivered along the imaging axis. On glass that is a specular-surface tool: it produces strong returns from flat polished faces and from the beam-splitter itself, either of which can appear as an extra edge.

The test is to disconnect the coaxial source and re-capture. If the extra edge disappears, the cause is the illumination path, and the remedy is to illuminate in transmission instead.

Where POMEAS optics land on those six

Reading the table against the six interception points:

- PMS-LTCM023-90 publishes the largest depth of field here at 7.25 mm, with 13.98 µm resolution and NA 0.024 at 0.23× and 90 mm — the model whose figures support thick glass with both faces in focus. - PMS-LTCM1-110 publishes 0.89 mm with 7.46 µm resolution and NA 0.045. That is small enough to isolate one plane in a thick part, which is what check 5 requires when the bore wall is the problem. - PMS-LTC2-110C and PMS-LTC1-110C are the coaxial options at 0.5 mm and 1.6 mm. Use them when the part is reflective rather than transmissive; on transmissive glass the coaxial path becomes a candidate cause instead. - PMS-LTFM05-179 is the F-mount option, publishing 18 µm resolution, a 72.0 mm × 48.0 mm field on a 35 mm sensor and F 13, with panel and glass edge work listed among its applications. It publishes no depth of field.

No NA or resolution figure here should be read as equivalent to another model's; blanks mean not published.

What those numbers mean on the line

Scenario 1 — Cover glass with a doubled edge on every hole

The station reports two edges per hole and a diameter that varies with lighting. The part is 1.1 mm thick and the lens publishes 0.89 mm of depth of field (PMS-LTCM1-110), so the faces are not both in focus: the station measures one sharp edge and one bloomed edge. Either move to a model with a larger depth of field, or keep the shallow one and focus deliberately on the entry face so the exit face never crosses the threshold.

Scenario 2 — Glass panel edge drifting with ambient light

A large panel is measured in reflection and the edge position moves through the shift. This is check 3: transmission is the correct geometry for a transparent part. Fitting a VP-GLTD collimated backlight with a beam diameter large enough to cover the field — 100 mm, 145 mm, 189 mm or 265 mm — converts the edge from a weak reflected feature into a silhouette step.

Scenario 3 — A bore wall producing a bright ring inside the hole

The apparent hole diameter is smaller than the gauge says and the inside of the hole is bright: the wall is guiding light. Collimated backlight reduces the grazing-angle component; if it does not, the measurement has to be pinned to a defined depth, which means choosing the depth of field deliberately.

Scenario 4 — An extra edge that disappears when the coaxial light is unplugged

A station running PMS-LTC2-110C on a polished glass component sees a faint second contour at a fixed offset, and disconnecting the coaxial source removes it. This is check 6 — the beam-splitter path, not the glass. Illuminating in transmission resolves it, and PMS-LTC2-110, the non-coaxial version at the same 2.0× and 110 mm, is the alternative.

What this article does not tell you

This is a technical reference, not a diagnosis of your station. No claim is made about what is causing ghost edges on a specific installation. Determining which surface a station is actually imaging requires access to the part and the images.

Optical performance is not measurement accuracy. Resolution and depth of field are properties of the lens; measurement accuracy additionally depends on pixel sampling, contrast, edge definition, illumination stability, and the algorithm. On transparent parts the edge definition term is usually the larger one, which is why this article puts weight on illumination geometry.

Every figure here is the published value for the specific POMEAS model named beside it, not measured under a single common condition. Where a field is absent — NA for PMS-LTC2-110C, PMS-LTC1-110C and PMS-LTFM05-179, depth of field for PMS-LTFM05-179 — read it as "not published".

The telecentric content already on this site covers sensor matching, numerical aperture and working distance selection as separate pages. This article is about the transparent-part case.

FAQ

Why does glass show more than one edge per feature?

Because every interface where the refractive index changes produces one: the entry face, the exit face, the bore wall and any chamfer, all at different depths and with different contrast.

Is a ghost edge an optical defect in the lens?

Usually not. A true ghost comes from a reflection between two surfaces, but most reported ghost edges are real images of a second surface the algorithm was never told to ignore.

Does a telecentric lens remove ghost edges on glass?

It removes the magnification change with depth, which is a large part of the problem on a conventional lens. It does not make the second surface disappear, and it does not change contrast.

Why is backlight better than ring light for transparent parts?

Reflection off glass produces weak edges and strong specular highlights. Transmission produces a silhouette with high contrast, and a collimated backlight additionally reduces the light that wraps around the edge.

How does depth of field affect a doubled edge on glass?

If the part is thicker than the depth of field, one face is defocused and its edge becomes a wide low-gradient feature that thresholds unreliably. PMS-LTCM1-110 publishes 0.89 mm; PMS-LTCM023-90 publishes 7.25 mm.

What wavelength should I use for glass?

It depends on the material and on which feature you need contrast for; the VP-GLTD series offers red, green, blue and white so the wavelength can be selected rather than assumed.

Next Step

If you are specifying a replacement and need measured values for your own conditions — your sensor size, your working distance, your feature size — send us those four numbers and we will run them against the current catalogue and tell you what actually fits, including when nothing does.