How to Detect Scratches on Glass?
Machine Vision Imaging Challenges and Solutions for Transparent Materials
How to Detect Scratches on Glass?
Summary
The key to glass scratch inspection is not simply increasing camera resolution, but building a complete imaging chain: lighting → lens → camera → exposure → trigger → image processing. Fine scratches, shallow scratches, pits, bubbles, and cracks must first form stable contrast in the image before algorithms can segment them reliably. For transparent and highly reflective materials, low-angle lighting, dark-field illumination, polarization, line scan, GigE Vision interfaces, and stable triggering are especially important.
1. Why Are Glass Surface Scratches So Difficult to Detect?
Glass looks smooth, clean, and transparent, but it is actually one of the more challenging materials for automated inspection. Fine scratches, shallow scratches, pits, bubbles, cracks, contamination, and edge defects are not simply "invisible" to the camera; rather, they fail to create sufficient image contrast against the glass background. The same scratch can produce completely different inspection results when the lighting, angle, or exposure changes. For glass, cover glass, optical glass, glass bottles, and other transparent materials, the real challenge is the complete imaging system, not just a high-resolution industrial camera. Similar challenges also exist in film inspection and semiconductor wafer inspection.
2. What Defects Can Be Found on Glass Surfaces?
Common defects include scratches, cracks, pits, bubbles, contamination, dents, edge chipping, foreign particles, and coating defects.
Scratches are often the most challenging because they can be extremely fine, with significant variation in orientation, depth, and width. Cracks appear as irregular linear structures; pits are small local defects; bubbles are circular structures inside or on the surface; contamination causes grayscale or color abnormalities; dents indicate local topographic changes; edge chipping is an edge structure abnormality; foreign particles may appear as bright or dark spots; coating defects change color or reflection characteristics.
3. The Biggest Challenge: Reflections
When light directly hits a glass surface, strong specular reflection can appear. If the camera is positioned within the reflection path, the image may contain overexposed areas, bright spots covering defects, glare, uneven backgrounds, low scratch contrast, and unstable inspection results at different positions. Therefore, brighter lighting does not necessarily mean better inspection. In glass inspection, lighting direction and incidence angle are often more important than simply increasing illumination intensity.
4. How Can We Make Scratches More Visible?
The basic idea is to change how light interacts with the glass surface so that scratches produce stronger intensity differences. Common methods include low-angle lighting, dark-field illumination, and backlighting.
Low-angle lighting directs light close to the glass surface. Scattering changes at the scratch, so it can appear as a bright or dark line under the right viewing angle. It is suitable for fine scratches, abrasions, linear defects, and small surface irregularities.
Dark-field illumination keeps the normal glass surface relatively dark while defects scatter light, creating a "dark background + bright defect" effect. It is very effective for fine scratches.
Backlighting is useful for glass outlines, edges, and certain internal defects. Its advantages include clear contours, easy edge segmentation, and stable backgrounds, making it friendly to dimensional inspection. However, backlighting is not automatically the best solution for all surface scratches. If a scratch is mainly visible because of surface scattering rather than contour changes, pure backlighting may not produce sufficient contrast.
5. Can Polarized Lighting Be Used for Glass Inspection?
Yes, depending on the defect type. Polarization can help control certain reflections and improve contrast for some surface defects. Adjustable factors include the polarization direction of the light source, a polarizing filter on the camera, illumination angle, and observation angle.
However, polarization is not a universal solution. Different glass materials, coatings, and surface treatments respond differently to polarized light. Testing with real samples is highly recommended before building a production line.
6. How to Choose a Camera for Glass Surface Inspection?
Many projects make a common mistake: "The scratch is very small, so choose the highest-resolution camera." High resolution is important, but it is only one part of the system. A suitable glass inspection solution must consider resolution, field of view, defect size, pixel resolution, motion speed, exposure time, interface bandwidth, and trigger requirements. Therefore, industrial camera selection and how to select an industrial camera should not rely on pixel count alone. The choice of industrial camera manufacturer also matters because SDK, interfaces, triggering, and after-sales support affect integration.
Special scenarios may evaluate: HDR camera for high-dynamic-range scenes; event camera for high-speed transient changes; embedded vision camera for embedded devices; near-infrared camera for near-infrared bands; LWIR thermal camera for long-wave infrared thermal imaging; LED flicker mitigation to avoid LED flicker affecting image stability.
Resolution calculation example: inspection width 100 mm, camera horizontal resolution 5000 pixels, then 100 mm ÷ 5000 = 0.02 mm/pixel = 20 μm/pixel. Theoretically, a 20 μm defect may form analyzable information. Actual limits also depend on optical resolution, MTF, motion, lighting, and algorithms.
7. Area Scan or Line Scan Camera?
It depends on the production process. Area scan cameras are suitable for individual glass sheets, cover glass, optical glass, flat glass, fixed stations, and robotic handling. Line scan cameras are better for continuous glass materials, inline inspection, large-format glass, continuous production lines, and high-speed moving materials. For continuous inline inspection, line scan vs area scan selection should be based on speed, width, defect size, resolution, data volume, and interface. Standardized interfaces such as GigE Vision are important in these systems.
8. How to Choose the Right Lens for Glass Inspection?
The lens is as important as the camera. Many projects have "clear images but poor scratch detection," and the problem may come from the lens. Lens selection should consider field of view, working distance, sensor size, focal length, resolution, distortion, depth of field, and installation space.
For general flat glass, an industrial lens can be selected based on FOV and working distance. For dimensional measurement, edge position, or geometric accuracy, telecentric optics may be considered. How to choose the right lens for an industrial camera cannot be answered by focal length alone.
9. Standard Lens vs Telecentric Lens
Standard lenses are cost-effective, widely available, and flexible for general inspection. Telecentric lenses are more suitable for precision measurement, edge inspection, dimensional measurement, and applications sensitive to perspective changes.
If the goal is simply "Is there a scratch?", address imaging contrast first. If the goal becomes "Is the edge dimension within tolerance?", re-evaluate the optical system. Telecentric lenses should not be used by default simply because the application is industrial.
10. Why Do Exposure, Gain, and ROI Matter?
Too long exposure causes motion blur, blurred edges, lower contrast, and lost detail. Too short exposure causes dark images, lower SNR, and difficulty distinguishing small defects. Exposure should be optimized with lighting intensity, motion speed, and sensor performance.
Gain is not always better when increased. High gain amplifies noise, causing false defects, stronger background texture, buried scratches, and unstable thresholds. Recommended sequence: lighting → lens → exposure → gain.
If defects only occur in specific regions, ROI can reduce irrelevant data, minimize background interference, and reduce processing workload.
11. A Typical Machine Vision Configuration for Glass Scratch Inspection
| Module | Recommended Approach |
|---|---|
| Industrial Camera | Select resolution based on defect size and FOV |
| Lens | Match FOV, WD, sensor size |
| Lighting | Prioritize testing low-angle, dark-field |
| Polarizer | Evaluate polarization for highly reflective scenarios |
| Trigger | External trigger recommended for high-speed lines |
| Software | ROI + filtering + segmentation + defect classification |
| Interface | GigE, USB3, high-speed interfaces, GigE Vision |
There is no universal configuration for all glass. Thickness, surface treatment, coating, color, speed, and defect type can all change the final design.
12. From Glass Inspection to Semiconductor and Film Inspection
The imaging methodology for glass inspection can be extended to film inspection and semiconductor wafer inspection. In semiconductor wafers, films, and transparent substrates, defects may also appear as low contrast, scattering changes, or reflection changes. Lighting angle, dark-field illumination, polarization control, line scan, GigE Vision interfaces, and stable triggering remain critical. For special bands, near-infrared camera and LWIR thermal camera solutions can be evaluated.
13. From Glass Inspection to Autonomous Driving and Automotive Perception
Machine vision is also widely used in autonomous driving and automotive perception. For how to choose a camera for autonomous driving, it is not enough to look at single-camera specifications; the reliability of a sensor fusion solution and an automotive-grade camera must also be considered. sensor fusion for self-driving cars requires integrating cameras, radar, LiDAR, and other sensors. night-time autonomous driving perception and all-weather perception place higher demands on dynamic range, low-light performance, and interference resistance. Specific scenarios include port autonomous trucking perception, roadside perception unit, AVP vision system, and in-cabin DMS. Related technologies also include HDR camera, LED flicker mitigation, and event camera.
14. From Glass Inspection to Pharmaceutical Inspection
The pharmaceutical industry is an important application area for machine vision. pharmaceutical inspection typically includes tablet inspection, capsule inspection, vial inspection, ampoule inspection, and blister packaging inspection. It needs to detect missing tablets, damage, foreign particles, fill level, sealing, labels, and packaging defects.
Similar to glass inspection, pharmaceutical inspection requires stable lighting, suitable lenses, reliable triggering, and traceable algorithms. For transparent vials, glass ampoules, and blister packaging, reflection control, low-angle illumination, and dark-field illumination are equally important.
15. From Glass Inspection to Medical Imaging and Life Science
In medical and life science fields, medical imaging camera, IVD camera, in vitro diagnostics camera, genetic testing camera, life science imaging, and ophthalmic imaging camera have higher requirements for image quality, sensitivity, noise, and stability.
Whether in in vitro diagnostics, genetic sequencing, microscopy imaging, or ophthalmic imaging, the core remains the complete imaging chain: sensor, lens, lighting, interface, triggering, and algorithms. The principle that "good vision starts with good imaging" also applies here.
16. Why Camera Specifications Alone Are Not Enough
Consider two projects using the same industrial camera. System A has proper lighting, matched lens, stable exposure, and accurate positioning, so results may be excellent. System B has strong reflection, mismatched FOV, excessive exposure, and position variation, so even a higher-resolution camera may produce many misses.
Therefore, pixel count determines how much spatial information can be represented, but it does not alone determine whether a defect is visible. This is especially true for transparent, highly reflective materials such as glass. When choosing an industrial camera manufacturer, consider whether it can provide complete imaging support, not just camera hardware.
17. From Seeing the Scratch to Detecting It Reliably
A complete glass inspection system can be divided into three stages. First, make the defect visible through lighting angle, dark-field, low-angle illumination, polarization, backlighting, and reflection control. Second, capture the defect with an appropriate sensor, sufficient resolution, a suitable lens, stable exposure, and reasonable gain. Third, classify the defect using ROI, thresholding, blob analysis, edge detection, pattern analysis, and defect classification.
Core principle: Good vision starts with good imaging.
18. FAQ: Common Questions on Glass Surface Scratch Inspection and Transparent Material Imaging
Human observers naturally change viewing angles, while a fixed industrial camera observes from a specific direction. The scratch may have very low grayscale contrast against the background. Optimize lighting, viewing angle, and exposure rather than simply increasing resolution.
Not necessarily. Required resolution depends on defect size, FOV, working distance, inspection accuracy, and material speed. Lens and lighting are equally important.
If defects mainly involve brightness or scattering, a monochrome camera often provides more stable grayscale contrast. If color, coating, or chromatic variation is involved, a color camera may be considered.
Fixed stations and individual sheets are often suitable for area scan. Continuous motion, large format, and high-speed inline inspection should evaluate line scan.
The problem may not be resolution. If lighting does not create sufficient contrast, a higher-resolution camera may simply produce a sharper version of the same low-contrast image.
Because glass is transparent and highly reflective. A scratch may be very fine and shallow. If lighting does not create stable contrast between the scratch and the background, even a higher-resolution HDR camera or high-speed industrial camera will only produce a "sharper low-contrast image." The core imaging chain is: lighting → lens → camera → exposure → trigger → image processing. Low-angle lighting, dark-field illumination, polarization control, and LED flicker mitigation are often more important than simply increasing pixel count.
For fine scratches, surface abrasions, linear defects, and small surface irregularities, evaluate low-angle lighting and dark-field illumination first. Dark-field illumination often creates a "dark background + bright defect" effect, which is very effective for fine scratches. Low-angle lighting may make scratches appear as bright or dark lines. For glass outlines, edges, dimensions, or certain internal defects, consider backlighting. Its advantages include clear contours, easy edge segmentation, and stable backgrounds. However, backlighting is not automatically the best solution for all surface scratches. If a scratch is mainly visible because of surface scattering rather than contour changes, pure backlighting may not produce sufficient contrast.
No. Polarization can control certain reflections and improve contrast for some surface defects, especially in highly reflective glass scenarios. It can reduce interfering reflections by adjusting the polarization direction of the light source, a polarizing filter on the camera, illumination angle, and observation angle. However, polarization is not a universal solution. Different glass materials, coatings, and surface treatments respond differently to polarized light. Before building a production line, testing with real samples is highly recommended.
Area scan cameras are suitable for individual glass sheets, cover glass, optical glass, flat glass, fixed inspection stations, and robotic handling. Line scan cameras are better for continuous glass materials, inline inspection, large-format glass, continuous production lines, and high-speed moving materials. For continuous inline inspection, evaluate line scan vs area scan camera selection carefully. Consider material speed, inspection width, defect size, required resolution, data volume, and interface. GigE Vision is a standardized industrial camera interface protocol that supports multi-vendor integration, long-distance transmission, and system expansion. For glass inline inspection, film inspection, and semiconductor wafer inspection, stable interfaces and triggering are equally important.
industrial camera selection and how to select an industrial camera should not rely on pixel count alone. Evaluate: 1. Defect size and required pixel resolution; 2. Field of view and working distance; 3. Motion speed and exposure time; 4. Interface bandwidth and transmission distance; 5. Triggering and synchronization requirements; 6. SDK, Linux/ARM Linux support, and after-sales service; 7. Whether the industrial camera manufacturer can provide complete imaging support, not just camera hardware. For glass inspection, the camera should be part of a complete vision system, not an isolated hardware component.
HDR camera is suitable for high-dynamic-range scenes, such as highly reflective glass inspection with strong brightness differences. event camera is suitable for high-speed transient changes, moving targets, and low-latency scenarios. embedded vision camera is suitable for embedded devices, compact vision systems, and edge computing. near-infrared camera is suitable for near-infrared imaging, which can be used in certain semiconductor, film, or special material inspection. LWIR thermal camera is suitable for long-wave infrared thermal imaging, used for temperature distribution, thermal defects, or special perception scenarios. In glass inspection, whether to use these cameras depends on defect characteristics, spectral response, motion speed, and system architecture.
In high-speed inspection, LED lighting may flicker. If camera exposure is not synchronized with LED flicker, image brightness may fluctuate, causing unstable defect contrast, false defects, or missed detection. LED flicker mitigation can reduce flicker by synchronized triggering, appropriate exposure time, constant-current driving, or high-frequency dimming, improving image consistency. This is important for glass scratch inspection, high-speed inline inspection, and autonomous driving perception.
All three involve transparent, semi-transparent, highly reflective, or low-contrast materials. Defects may appear as scattering changes, reflection changes, grayscale abnormalities, or tiny topographic changes. Common keys include low-angle lighting, dark-field illumination, polarization control, line scan, stable triggering, GigE Vision interfaces, and high-resolution lenses. For film inspection and semiconductor wafer inspection, special spectral bands such as near-infrared camera, LWIR thermal camera, or SWIR solutions may also be considered.
how to choose a camera for autonomous driving should not rely only on single-camera specifications. It must also consider automotive-grade reliability, temperature range, vibration, functional safety, dynamic range, and interfaces. An automotive-grade camera must meet automotive-grade requirements and operate stably in complex environments. sensor fusion solution and sensor fusion for self-driving cars emphasize integrating cameras, radar, LiDAR, IMU, GNSS, and other sensors to improve perception reliability. In these systems, HDR camera, LED flicker mitigation, and event camera are also often discussed.
night-time autonomous driving perception: Night-time autonomous driving perception, emphasizing low light, glare, and dynamic range. all-weather perception: All-weather perception, emphasizing rain, snow, fog, strong light, and backlight. port autonomous trucking perception: Port autonomous trucking perception, involving containers, lifting, complex traffic, and precise positioning. roadside perception unit: Roadside perception unit for vehicle-road cooperation, traffic monitoring, and edge computing. AVP vision system: Automated valet parking vision system for low-speed parking and obstacle recognition. DMS: Driver monitoring system for fatigue, distraction, gaze, and posture monitoring. These applications rely on stable imaging, reliable triggering, HDR camera, LED flicker mitigation, and sensor fusion solution.
pharmaceutical inspection is a general term for pharmaceutical vision inspection, typically including: tablet inspection: Missing tablets, cracks, color differences, stains, and dimensions. capsule inspection: Capsule damage, deformation, color, and foreign particles. vial inspection: Foreign particles, fill level, sealing, cracks, and neck defects. ampoule inspection: Cracks, foreign particles, fill level, and sealing. blister packaging inspection: Missing pills, damage, sealing, printing, and packaging defects. Similar to glass inspection, these applications require stable lighting, suitable lenses, reliable triggering, and traceable algorithms. For transparent vials, glass ampoules, and blister packaging, reflection control, low-angle illumination, and dark-field illumination are equally important.
These medical and life science applications typically require: High sensitivity, low noise, high dynamic range, stable color or grayscale consistency, long-term stable operation, reliable interfaces and SDK, and integration into microscopes, diagnostic devices, or ophthalmic equipment. medical imaging camera, IVD camera, in vitro diagnostics camera, genetic testing camera, life science imaging, and ophthalmic imaging camera all emphasize the complete imaging chain: sensor, lens, lighting, interface, triggering, and algorithms. The principle that "good vision starts with good imaging" also applies to these high-precision medical imaging applications.
A complete glass inspection system can be divided into three stages: 1. Make the defect visible: lighting angle, dark-field, low-angle illumination, polarization, backlighting, reflection control. 2. Capture the defect: appropriate sensor, sufficient resolution, suitable lens, stable exposure, reasonable gain. 3. Classify the defect: ROI, thresholding, blob analysis, edge detection, pattern analysis, defect classification. Core principle: Good vision starts with good imaging.
Check in this order: 1. Lighting: Are low-angle, dark-field, or polarization methods used? Is there overexposure or reflection? 2. Lens: Do FOV, WD, sensor size, resolution, and distortion match? 3. Exposure: Is there motion blur or underexposure? Is SNR sufficient? 4. Gain: Is gain too high, causing noise to bury fine scratches? 5. Trigger and motion: Are they synchronized? Is there motion blur? 6. ROI and algorithm: Is the correct region processed? Are thresholds stable? 7. Camera: Finally evaluate whether an HDR camera, event camera, embedded vision camera, near-infrared camera, or LWIR thermal camera is needed. If lighting does not create sufficient contrast, a higher-resolution camera will only produce a sharper version of the same low-contrast image.
19. Conclusion
Glass scratch inspection is not simply a matter of selecting a high-resolution industrial camera. The complete imaging chain determines performance: defect → lighting → lens → camera → exposure → trigger → image processing. Lighting is particularly important.
For fine scratches, evaluate low-angle and dark-field illumination first. For highly reflective surfaces, consider polarization. For continuous high-speed production, evaluate line scan cameras, exposure time, triggering, and data transmission together.
Camera selection should be based on actual FOV, defect size, material speed, resolution, interface, and system architecture, not simply the highest pixel count. A camera for glass surface inspection should be part of a complete machine vision system, not an isolated hardware component.
20. MindVision Industrial Cameras for Glass Inspection
MindVision can provide industrial cameras and customized imaging solutions based on field of view, speed, resolution, interface, and integration requirements for glass inspection projects. Area scan cameras can be used for fixed stations; line scan cameras can be considered for continuous high-speed inspection; GigE, USB3, high-speed interfaces, or GigE Vision solutions can be evaluated for high-speed transmission.
Automation integration may include external triggering, programmable I/O, flash synchronization, multi-camera systems, embedded integration, Linux/ARM Linux environments, and SDK development. The product portfolio covers GigE, USB3, CoaXPress, and board-level cameras/modules. For non-standard vision applications such as glass inspection, the key is to build a complete imaging system: Sensor + Lens + Lighting + Interface + Trigger + Software.
#MindVision
MindVision Technology Co., Ltd. is a professional domestic industrial camera manufacturer headquartered in Shenzhen, with production centers and R&D bases in Changsha (Hunan) and Wuxi (Jiangsu), employing nearly 300 people. Its product portfolio covers area-scan cameras, line-scan cameras, thermal imaging cameras, SWIR cameras, UV cameras, and stereo camera modules, with over 100 patents and software copyrights. The company is ISO9001 certified.
For more information or free selection consulting, please visit MindVision's official website or contact the sales team.
We’ll be glad to help you
Please Login
Full Name
Company Name
E-mail*
Country / Region
Your Needs / Application*
Verification code
Get in touch
Name*
E-mail*
Country*
Your Needs / Application*
