A Review of Surface Defect Detection in Machine Vision

Surface defects in industrial products adversely affect their aesthetics, comfort, and functional performance; therefore, manufacturers inspect these defects to detect and control them promptly.

Machine-vision-based inspection methods can largely overcome the shortcomings of manual inspection—such as low sampling rates, limited accuracy, poor real-time performance, low efficiency, and high labor intensity—and are therefore receiving increasingly widespread research and application in modern industry.

This paper takes machine-vision-based surface defect detection as its research subject. Based on an extensive review of relevant literature and recent advances, it provides a comprehensive overview of the applications of machine vision in this field. The work analyzes the operating principles and basic architecture of typical machine-vision systems for surface defect inspection, discusses the current state of research, existing visual software and hardware platforms, and reviews related theoretical and algorithmic developments, including image preprocessing techniques, image segmentation methods, image feature extraction and selection algorithms, and pattern recognition approaches. Furthermore, it summarizes the fundamental concepts, salient characteristics, and inherent limitations of each major approach, while offering on potential future directions of development.

In machine vision-based surface defect detection systems, image processing and analysis algorithms constitute a critical component, each with its own strengths, weaknesses, and applicable domains. Enhancing the accuracy, real-time performance, and robustness of these algorithms has long been a central focus of research.

Conclusion: Machine vision is an emulation of human vision, and surface inspection in machine vision encompasses numerous disciplines and theoretical frameworks. Further advancing this field toward greater automation and intelligence will require more in-depth research.

Surface Defect Detection

Machine vision technology is a non-contact, non-destructive automated inspection technique and an effective means of achieving equipment automation, intelligence, and precision control. It boasts notable advantages such as safety and reliability, a broad spectral response range, the ability to operate continuously in harsh environments, and high production efficiency. A machine vision inspection system acquires surface images of products using appropriate lighting and an image sensor (CCD camera), extracts feature information from these images via suitable image-processing algorithms, and then performs operations such as defect localization, identification, and classification, as well as statistical analysis, data storage, and query functions, based on this feature information.

Basic Components of a Machine Vision Surface Defect Detection System

It mainly comprises an image acquisition module, an image processing module, an image analysis module, and a data management and human–machine interface module.

The image acquisition module consists of an industrial camera, an optical lens, a light source, and a mounting fixture, among other components, and its function is to capture images of the product’s surface. Under illumination from the light source, the optical lens projects an image of the product’s surface onto the camera sensor; the light signal is first converted into an electrical signal and then further transformed into a digital signal that can be processed by a computer. Currently, industrial cameras are predominantly based on CCD or CMOS sensors. Among these, CCD remains the most widely used image sensor in machine vision.

Machine vision illumination directly affects image quality by mitigating ambient light interference, ensuring image stability, and maximizing contrast. Commonly used light sources today include halogen lamps, fluorescent lamps, and light-emitting diodes (LEDs). LED lighting has gained widespread adoption thanks to its compact size, low power consumption, fast response time, excellent monochromaticity, high reliability, uniform and stable output, and ease of integration.

Lighting systems composed of light sources can be classified, according to their illumination method, into bright-field illumination and dark-field illumination, as well as structured-light illumination and strobe illumination. Bright-field and dark-field primarily describe the spatial relationship between the camera and the light source: bright-field illumination refers to the camera directly capturing the reflected light from the target, with the camera typically positioned on the opposite side of the light source—this configuration is easy to set up; dark-field illumination, by contrast, involves the camera indirectly detecting scattered light from the target, with the camera usually located on the same side as the light source, offering the advantage of producing high‑contrast images. Structured-light illumination projects a grating or a line‑type light source onto the object under inspection, and by analyzing the resulting distortions, it extracts three-dimensional information about the object. Strobe illumination uses high‑frequency light pulses directed at the object, with the camera’s exposure synchronized to the light source.

The image processing module primarily encompasses image denoising, image enhancement and restoration, defect detection, and object segmentation. Due to factors such as the on-site environment, the photoelectric conversion of CCD images, transmission circuits, and electronic components, images inevitably acquire noise, which degrades image quality and adversely affects subsequent processing and analysis. Therefore, preprocessing is required to remove this noise.  Image enhancement aims to deliberately emphasize the overall or local characteristics of an image according to its intended application, making previously unclear images sharper or highlighting specific features of interest, while enhancing the distinctions among different objects in the image and suppressing irrelevant details. This process improves image quality, enriches information content, and enhances the effectiveness of visual interpretation and recognition.  Image restoration is a computational procedure that reconstructs or recovers images whose quality has deteriorated. In many cases, restoration employs techniques similar to those used for enhancement; however, the results of enhancement typically require validation in a subsequent stage. By contrast, image restoration seeks to leverage prior knowledge of the degradation process to recover the original appearance of a degraded image, such as eliminating additive noise or reversing motion blur.  The goal of image segmentation is to isolate target regions within an image, facilitating subsequent processing steps.

Surface Defect Detection Applications

The applications are extremely broad, encompassing such fields as steel and iron metallurgy, nonferrous metal processing, high-precision copper sheet and strip, aluminum sheet and strip, aluminum foil, stainless steel manufacturing, electronic materials, nonwoven fabrics, textiles, glass, paper, and thin films.

Why use a surface defect detection system?

Ensure product quality, improve production processes, and reduce labor costs.

The main components of the line-scan surface defect detection system are:

The vision acquisition subsystem primarily comprises a line-scan camera, lens, light source, and image acquisition card.

The system bracket assembly comprises a camera bracket, a light source bracket, and a console bracket.

The electrical section (communication/control unit) includes encoders, motion control cards or PLCs, and may also comprise motors, among other components.

Others: various wires and cables, CL‑type cables, power cords, assorted SMPS units, lighting controllers, and so on.

There’s also PC.