Why can’t surface scratches be fixed simply by shining light on them?

This case is quite typical. While the issue appears to be an algorithmic one on the surface, the root cause often lies in the lighting setup at the front end.

In many on-site debugging sessions, the bottleneck isn't the algorithm itself—it's the very first image. If the target doesn't appear in the image, subsequent adjustments to thresholds, filters, and models will become extremely difficult.

Without getting into complicated formulas, let’s look at it the way we’d approach it on a construction site: Where was the original image unstable, and where does the new lighting make it more stable? Also, what are the limitations of this solution?

Case 002: Surface Scratch Detection

This case is testing...Surface Scratch DetectionA sample can be simply understood as...Membership card, material/surface properties can be categorized asPlastics/Composite Materials.

The most troublesome point on site is:Background interference, insufficient contrastThe effective solution is:Parallel Coaxial Brightfield Illumination.

The biggest problem with images of this type isn't darkness—it's the excessive amount of irrelevant background. First, suppress the background; it’s more reliable to do this than to rely on algorithmic enhancements afterward.

Parallel coaxial brightfield microscopy places greater emphasis on directional consistency, making it ideal for suppressing background and observing shallow features.

The last thing to look at isn't how beautiful the parameters are, but whether the results have become more stable: highlighting the scratches while eliminating background interference.

Such schemes also have limitations: Coaxial illumination works effectively for mirror-like and flat samples, but may produce uneven illumination on strongly curved surfaces or samples with significant height variations.

Case 050: Scratch Detection on Cover Glass

This case is testing...Cover Glass Scratch DetectionA sample can be simply understood as...Samsung cover glass, material/surface properties can be categorized asGlass/Transparent Materials.

The most troublesome point on site is:Insufficient contrastThe effective solution is:Multi-angle line lightDark-field illumination.

When contrast is insufficient, simply increasing brightness may not necessarily help. The key is to create a stable grayscale difference between the target and the background.

Multi-angle brightfield and darkfield illumination is suitable for scratches with irregular orientations, preventing illumination from being limited to just one particular direction.

The last thing to look at isn't how beautiful the parameters are, but whether the results have become more stable—highlighting the scratch contours.

Such solutions also have limitations: they require real-world verification, taking into account the sample material, target size, and camera angle.

Case 054: Copper Pipe Head Scratch Detection

This case is testing...Copper Pipe Head Scratch DetectionA sample can be simply understood as...Copper pipe, material/surface properties can be categorized asMetal/high-reflection material, cylindrical or curved surface structure.

The most troublesome point on site is:Reflection interferenceThe effective solution is:High-angle brightfield illumination.

The issue of reflections cannot be solved simply by increasing exposure; the key lies in controlling the proportion of direct reflections that enter the lens.

High-angle brightfield illumination can stably illuminate the main surface, but be mindful of specular reflections from highly reflective materials.

The last thing to look at isn't how beautiful the parameters are, but whether the results have become more stable: highlight the scratches and avoid interference from surface textures.

Such schemes also have limitations: they require real-world verification, taking into account the sample material, target size, and camera angle.

How do you judge it on the spot?

If you encounter a similar issue, I don’t recommend immediately asking, “What’s the best light source to use?” A more practical way to ask would be:

1. Is the target now separated from the background?

2. After the light is changed, does it enhance only the target, or does it also enhance irrelevant textures?

3. Can this image be consistently reproduced, rather than looking good only on a specific sample?

The value of a lighting setup isn't to make the images look prettier—it's to reduce the amount of guesswork the algorithm has to do.

Summarize in one sentence.

Why can’t surface scratches be addressed simply by increasing illumination? The key isn’t merely boosting brightness; rather, it’s about creating stable imaging differences.

The image is very bright but the detection is unstable—this kind of situation is all too common on-site. What really needs to be addressed is: Are the target features clearly highlighted, and have the interfering signals been effectively suppressed?