For automatic visual inspection of patterns on printed wiring boards and/or patterned wafers, this paper presents a new defect detection method for grayscale images without precise image alignment. Most of the conventional visual inspection algorithms based on grayscale reference comparison require precise image alignment with precision of subpixel or within +1 pixel; however, it is difficult to succeed the precise image alignment in every image. While a defect inspection method without precise image alignment has been previously proposed for binary images, the expansion to grayscale images we discuss is indispensable for detecting more minute defects. We propose dynamic tolerance control based on grayscale morphology to reduce false defects on pattern edges, and use gray dilation operation so that a weakness of the original method for binary images, an inability to detect the absence of minute patterns, is overcome. Theoretical analysis and experimental results show that the proposed method is capable of detecting subpixel-sized defects, and has practical detection performance.
展开▼