| Summary: | 碩士 === 元智大學 === 工業工程與管理學系 === 105 === For defect detection in non-periodically patterned images, such as printed-circuit board (PCB) or integrated circuit (IC) die found in the electronic industry, template matching techniques are popularly used to solve the problem. The sum of squared differences (SSD), sum of absolute differences (SAD), and normalization cross correlation (NCC) are commonly-used similarity measures to evaluate the anomaly in the two compared windows. The traditional template matching methods are sensitive to translation and illumination. Even under a well-aligned condition, the product variation in the manufacturing process is inevitable, and can cause false alarms in the inspection. To prevent the false alarm in the test image, the defect to be detected generally must contain at least a few pixels in width so that the true defect can be distinguished from the component edges using the traditional methods.
In this study, we propose a global template matching in the spectral domain using the Fourier transform. It is based on the comparison of the whole Fourier spectra between the template and the test image. It retains only the suspicious frequency components in the Fourier spectrum of the test image, and discards the common frequency components found in both Fourier spectra. The inverse Fourier transform is then applied to restore the test image, where the local anomalies will be preserved and the common background pattern will be removed as a uniform surface. A simple statistical control limit is finally used as the threshold to segment the local defect. The proposed spectral method is invariant to translation and illumination, and can detect subtle defects as small as one-pixel wide. A wide variety of PCB and IC images have been evaluated in the experiment, and the test results show that the proposed method is very effective and practically efficient for defect inspection in non-periodically patterned surfaces.
|
|---|
