Linear Defect Detection on Glass Substrate Capitalizing on Anisotropic Filters

The industrial production process necessitates stringent standards in quality and quantity for efficient process control. The diminished contrast of linear defects, such as scratches, in glass products destined for display terminals poses a significant challenge in defect detection. This paper presents a method developed on the Halcon platform, capitalizing on a set of linear convolution kernels ranging from 0 to 170° By applying regional grayscale thresholding and targeted anisotropic linear sharpening to images captured by a vision system, we significantly enhance the grayscale contrast between defective and non-defective areas. This approach markedly improves the detection rate of linear defects. Our method serves as a valuable case study for identifying low-contrast linear defects in industrial products.