Skip to main content

AI & machine vision:

Optical  properties

In today’s part, the focus is on the question of which optical properties the relevant features exhibit and which illumination strategies this implies concretely for a machine vision system.

Why optical properties are so critical

The way a feature reflects, scatters, absorbs, or transmits light directly determines whether it appears as contrast in the image or disappears in noise and reflections.

AI models are unforgiving here: if the feature is not cleanly present as a signal in the image, the model cannot reliably distinguish it.

Aufbau mit 4 Kameras die herabfallende Strassteine aufnehmen

At a high level, features and the base material can be grouped into the following classes:

  • Crack-/edge-like structures on smooth, often glossy surfaces
  • Residues, contamination, and films on specular or coated parts
  • Embossed, raised, or recessed structures (engravings, scribe marks)
  • Transmission features in transparent or translucent materials

For each of these classes, there are typical illumination concepts that are well supported by the underlying physics.

Cracks, scratches, and micro-defects: darkfield and low angle

Cracks, scratches, burrs, or fine edges on polished surfaces generally benefit from darkfield or low-angle illumination.

The principle: the part is illuminated at a very shallow incidence angle, most of the light is reflected away from the camera by the smooth surface, so the surface appears nearly dark.

Anomalies such as scratches, dents, or particles disturb the reflection pattern, scatter light toward the camera, and therefore appear bright on a dark background.

Typical applications include:

  • Scratch detection on metal surfaces and hard disks
  • Detection of tool marks, burrs, and indentations
  • Detection of particles, dust, and fingerprints on polished surfaces

Geometry is crucial:

  • Very shallow incidence angle (often below 10°)
  • Small working distances (typically a few millimeters to a few centimeters)
  • Mechanically stable mounting, since even small angle changes can significantly affect contrast.
Aufnahmen fallender Strasssteine aus vier Perspektiven

Residues on glossy surfaces: diffuse vs. coaxial

Residues, films, or contamination on specular or coated surfaces behave optically differently from sharp-edged cracks. The goal is often to control the specular base reflection so that residues become visible as intensity or texture changes.

Two typical approaches:

  • Diffuse illumination (e.g. dome light, tube light)
    • Produces very uniform lighting and reduces harsh reflections and shadows.
    • Well suited for slightly curved, glossy surfaces (e.g. body parts, plastic covers), where local changes in reflection caused by residues show up as deviations in an otherwise homogeneous image.
  • Coaxial (on-axis) illumination
    • Light is coupled in along the optical axis via a beam splitter and directed normally onto the surface.
    • Smooth, specular surfaces reflect the light directly back into the camera and appear uniformly bright; disturbances of the surface (residues, texture, microscopic roughness) alter the reflection and create contrast.
    • Particularly useful for planar, highly reflective surfaces such as wafers, metallized films, or glass.

    Example: soot or oil films on a highly polished part can often be imaged with a diffuse dome such that the base surface appears uniform, while the film regions exhibit slightly different intensity or texture.

    Coaxial illumination, on the other hand, is helpful when the surface is nominally perfectly specular and even small angular deviations or residues are to be emphasized via the directed reflection.

    Embossed and topographic features

    For engravings, scribe marks, embossed codes, or raised structures, two questions arise:

    • Do we primarily want to emphasize the edge/topography?
    • Or is the main goal to maximize readability of the code (e.g. 2D code, laser marking)?

    Darkfield ring lights are very well suited to highlight the edges of engravings or embossed structures. Due to the shallow incidence angle, edges appear as bright lines, while the base surface remains dark – ideal for detecting scribe marks on wafers or embossed numbers on metal parts.

    If the main requirement is contrast between marking and background (for example, in the case of laser-marked codes),

    • brightfield illumination
    • or coaxial systems
      can be the better choice, transforming differences in roughness and reflection behavior into intensity contrast.

    Transparent and translucent features

    In transparent or translucent materials (glass, films, plastic, pharmaceutical packaging), transmission plays a dominant role.

    Typical examples include:

    • Inclusions, bubbles, or scratches in glass or plastic
    • Misalignment or thickness variations in films
    • Fill level, particles, or air bubbles in transparent containers

    In these cases, other types of illumination are used:

    • Transmitted light / backlight: features become visible as absorption or refraction artifacts, contours appear with high contrast.
    • Structured illumination (e.g. line patterns): distortion of the pattern at defects or thickness variations can be used for evaluation and 3D reconstruction.
    • Coaxial illumination on translucent materials: reduces the contribution from volume scattering and emphasizes the surface, improving visibility of fine surface structures or shallow topography.

    In combination with AI, subtle patterns (e.g. haze, inhomogeneities) can be captured that are hard to robustly threshold with classical methods – but only if the illumination setup is stable and reproducible.

    Matte, diffuse surfaces and low contrast

    For highly diffuse, matte surfaces, specular reflection is minimal. The brightness distribution is primarily determined by the local reflectance properties.

    Challenges in this context include:

    • Very low contrast between the feature and the base material.
    • Texture noise (e.g., cast structures, matte coatings)

    Suitable approaches for such cases include:

    • Homogeneous brightfield area illumination with careful alignment to maximize global contrast.
    • Polarization filters on both the illumination and the optics to reduce residual reflections and suppress disturbing highlights.
    • Depending on the geometry, slight oblique illumination to subtly translate surface relief into brightness gradients without shifting into an extreme dark-field configuration.

    A typical example is the detection of slightly discolored or otherwise altered areas on a matte, textured surface (e.g., powder coating).

    In such cases, the objective is less about emphasizing sharp edges and more about maximizing the global signal-to-noise ratio (SNR) for subtle grayscale differences. This is an area where AI models can perform particularly well, provided that the lighting setup minimizes environmental variability.

    What was your experience with different illumination scenarios and AI ?